TorchData.ipynb

{
 "cells": [
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Learning TorchData Through Examples\n",
    "\n",
    "What follows is really more of a tutorial aggregator. I've tried to pull together the best pieces I've found from several TorchData tutorials and code snippets, and reorganize them all into one place. The goal of this article is simply to provide a good collection of information for anyone exploring TorchData. \n",
    "\n",
    "Note that in many cases *I've copied directly from the source text*, in some cases with minor edits, with the notion that the descriptions provided by the authors' of these libraries are already incredibly succinct and no benefit would be had by me writing them. I've provided links for sections that have been copied from other tutorials, and listed all references at the bottom.\n",
    "\n",
    "<a id=\"0\"></a> <br>\n",
    "## Table of Contents\n",
    "1. [Introduction to TorchData](#1)\n",
    "1. [Installation](#5)\n",
    "    1. [Prerequisites](#6)\n",
    "    1. [Installing](#7)\n",
    "1. [DataPipe Basics](#9)\n",
    "    1. [Using DataPipes](#10)\n",
    "    1. [Class constructor vs. Functional form](#11)\n",
    "    1. [IterDataPipes vs. MapDataPipes](#12)\n",
    "1. [Tabular Data Examples](#13)\n",
    "    1. [Using a DataPipe with DataLoader](#14)\n",
    "    1. [Multiprocessing](#15)\n",
    "    1. [DataLoader vs. DataLoader2](#16)\n",
    "    1. [Splitting data](#17)\n",
    "    1. [Computing number of samples](#18)\n",
    "    1. [TorchArrow](#19)\n",
    "1. [Vision Data Examples](#20)\n",
    "    1. [Torchvision](#21)\n",
    "    1. [Image Folder](#22)\n",
    "        1. [Visualizing DataPipes](#23)\n",
    "    1. [WebDataset](#24)\n",
    "        1. [Creating WebDataset data](#25)\n",
    "        1. [Using the WebDataset DatPipe](#26)\n",
    "        1. [Speeding up transforms with Kornia](#27)\n",
    "    1. [Object detection example](#28)\n",
    "1. [Language Data Examples](#29)\n",
    "    1. [TorchText](#30)\n",
    "    1. [HuggingFace](#31)\n",
    "1. [Custom DataPipes](#32)\n",
    "1. [Distributed dataloading](#33)\n",
    "    1. [Running on SageMaker](#34)\n",
    "1. [S3FileLoader](#35)\n",
    "1. [AIStore on S3](#36)\n",
    "1. [References](#37)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----\n",
    "\n",
    "## Introduction to TorchData <a class=\"anchor\" id=\"1\"></a>"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Hey there! Do you want to learn about the library called `torchdata`? It's a collection of data loading tools that make it easy to build fast and flexible pipelines. It's currently in beta.\n",
    "\n",
    "One of the cool things about torchdata is that it uses composable building blocks called DataPipes. A `DataPipe` is designed to work well with PyTorch's `DataLoader` and have a lot of useful features like loading files, parsing, caching, transforming, and more. [[1]](https://github.com/pytorch/data/blob/main/README.md#torchdata)\n",
    "\n",
    "In the past, people have had trouble using PyTorch's DataLoader because it had too many features combined in one place, making it difficult to extend or replace. And lots of libraries were reinventing the wheel by writing their own data loading tools. With torchdata, you can save time and effort by using these composable DataPipes instead. [[2]](https://github.com/pytorch/data/blob/main/README.md#why-composable-data-loading)\n",
    "\n",
    "Another reason why torchdata was created is to make sure that data loading doesn't become a bottleneck. With GPUs getting faster and faster, I/O and memory are becoming more likely to slow things down. By using these DataPipes, you can make sure that the GPU never has to wait for a batch of new data.\n",
    "\n",
    "DataPipes are a rebranding of PyTorch's Dataset, but with a focus on composability. They're just functions that accept an access function over Python data structures and return a new access function with a slight transformation applied. For example, you could have a `JsonParser` DataPipe that takes in a file name and raw stream and returns the file name and the deserialized data. [[3]](https://github.com/pytorch/data/blob/main/README.md#what-are-datapipes)\n",
    "\n",
    "With DataPipes, you can chain them together to create complex pipelines. And because they're designed to work with streamed data, you can process data one piece at a time without having to wait for all of it to be loaded into memory. For now, most of the focus is on `IterDataPipes`, but more support for `MapDataPipes` will come later.\n"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----\n",
    "\n",
    "## Installation <a class=\"anchor\" id=\"5\"></a>\n",
    "\n",
    "### Prerequisites <a class=\"anchor\" id=\"6\"></a>\n",
    "\n",
    "This code was tested using CUDA 11.6. To install a particular version of CUDA, your NVIDIA driver's compatibility version must be greater or equal to the CUDA version (top-right in `nvidia-smi`).\n",
    "\n",
    "Most of the setup can be accomplished in a few simple commands described below. A few examples require a [Kaggle](https://www.kaggle.com) account and [API key](https://www.kaggle.com/docs/api#authentication) to run a dataset download command."
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Installing <a class=\"anchor\" id=\"7\"></a>"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The libraries used in this tutorial can be installed by creating a Conda environment using the provided `environment.yml`:\n",
    "\n",
    "\n",
    "```bash\n",
    "    # Install mamba for faster conda installs\n",
    "    conda install mamba -n base -c conda-forge\n",
    "\n",
    "    # Create the environment with mamba\n",
    "    mamba env create --name torchdata --file=environment.yml\n",
    "```\n",
    "\n",
    "If the above installation was successful, you should be able to select `torchdata` as your Jupyter kernel."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[94mname\u001b[39;49;00m:\u001b[37m \u001b[39;49;00mtorchdata\u001b[37m\u001b[39;49;00m\n",
      "\u001b[94mchannels\u001b[39;49;00m:\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mpytorch-nightly\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mconda-forge\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mdefaults\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mnvidia\u001b[37m\u001b[39;49;00m\n",
      "\u001b[94mdependencies\u001b[39;49;00m:\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mpython=3.8.0\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mipython=8.4.0\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mpytorch=2.0.*\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mtorchvision=0.15.*\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mpytorch-cuda=11.6\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mtorchdata=0.6.*\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mtorchtext=0.15.*\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mlibjpeg-turbo=2.1.*\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mipykernel=6.19.*\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mkaggle=1.5.*\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00msvn=1.10.*\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mnb_conda_kernels=2.3.*\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mpip\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m  \u001b[39;49;00m-\u001b[37m \u001b[39;49;00m\u001b[94mpip\u001b[39;49;00m:\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m    \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mhttps://download.pytorch.org/whl/nightly/torcharrow-0.2.0a0.dev20230201-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m    \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mtqdm==4.64.1\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m    \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mwebdataset==0.2.31\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m    \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mdill==0.3.6\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m    \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mpandas==1.3.5\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m    \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mgraphviz==0.20.1\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m    \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mkornia==0.6.9\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m    \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mipywidgets==7.7.2\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m    \u001b[39;49;00m-\u001b[37m \u001b[39;49;00maistore==1.0.8\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m    \u001b[39;49;00m-\u001b[37m \u001b[39;49;00maistore[botocore]\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m    \u001b[39;49;00m-\u001b[37m \u001b[39;49;00msagemaker==2.129.0\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m    \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mdatasets==2.9.0\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m    \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mpycocotools==2.0.6\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m    \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mpyarrow==11.0.0\u001b[37m\u001b[39;49;00m\n",
      "\u001b[37m    \u001b[39;49;00m-\u001b[37m \u001b[39;49;00mpillow-simd==9.0.0.post1\u001b[37m\u001b[39;49;00m\n"
     ]
    }
   ],
   "source": [
    "!pygmentize environment.yml"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Verifying installation <a class=\"anchor\" id=\"8\"></a>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import sys\n",
    "from pkg_resources import parse_version as pv\n",
    "\n",
    "import torch\n",
    "import torcharrow\n",
    "import torchtext\n",
    "import torchvision\n",
    "import torchdata\n",
    "\n",
    "assert sys.version_info == (3, 8, 0, 'final', 0), \"Use Python 3.8.0\"\n",
    "assert torch.cuda.is_available(), \"This notebook requires GPU support.\"\n",
    "assert torch.version.cuda == \"11.6\", \"This notebook was tested with CUDA 11.6\"\n",
    "assert pv(torch.__version__) >= pv(\"2.0.0.dev20230201\")\n",
    "assert pv(torchvision.__version__) >= pv(\"0.15.0.dev20230201\")\n",
    "assert pv(torchtext.__version__) >= pv(\"0.15.0.dev20230201\")\n",
    "assert pv(torchdata.__version__) >= pv(\"0.6.0.dev20230201\")\n",
    "assert pv(torcharrow.__version__) >= pv(\"0.2.0a0.dev20230201\")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----\n",
    "\n",
    "## DataPipe Basics <a class=\"anchor\" id=\"9\"></a>\n",
    "\n",
    "The main idea is that features previously included in DataLoader, such as shuffling and batch size, will be transferred to DataPipes. Currently, `DataPipe` is compatible with `DataLoader` and can be used as a substitute for `Dataset`. In the future, `DataLoader2` will be used as a streamlined dataloader specifically designed for use with DataPipes."
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "### Using DataPipes [[4]](https://pytorch.org/data/main/dp_tutorial.html) <a class=\"anchor\" id=\"10\"></a>\n",
    "\n",
    "<blockquote>\n",
    "\n",
    "Suppose that we want to load data from CSV files with the following steps:\n",
    "\n",
    "- List all CSV files in a directory\n",
    "- Load CSV files\n",
    "- Parse CSV file and yield rows\n",
    "- Split our dataset into training and validation set\n",
    "\n",
    "There are a few [built-in DataPipes](https://pytorch.org/data/main/torchdata.datapipes.iter.html) that can help us with the above operations.\n",
    "\n",
    "- `FileLister` - [lists out files in a directory](https://pytorch.org/data/main/generated/torchdata.datapipes.iter.FileLister.html)\n",
    "- `Filter` - [filters the elements in DataPipe based on a given function](https://pytorch.org/data/beta/generated/torchdata.datapipes.iter.Filter.html)\n",
    "- `FileOpener` - [consumes file paths and returns opened file streams](https://pytorch.org/data/beta/generated/torchdata.datapipes.iter.FileOpener.html)\n",
    "- `CSVParser` - [consumes file streams, parses the CSV contents, and returns one parsed line at a time](https://pytorch.org/data/beta/generated/torchdata.datapipes.iter.CSVParser.html)\n",
    "- `RandomSplitter` - [randomly split samples from a source DataPipe into groups](https://pytorch.org/data/beta/generated/torchdata.datapipes.iter.RandomSplitter.html)\n",
    "\n",
    "As an example, the source code for `CSVParser` looks something like this:\n",
    "\n",
    "```python\n",
    "    @functional_datapipe(\"parse_csv\")\n",
    "    class CSVParserIterDataPipe(IterDataPipe):\n",
    "        def __init__(self, dp, **fmtparams) -> None:\n",
    "            self.dp = dp\n",
    "            self.fmtparams = fmtparams\n",
    "\n",
    "        def __iter__(self) -> Iterator[Union[Str_Or_Bytes, Tuple[str, Str_Or_Bytes]]]:\n",
    "            for path, file in self.source_datapipe:\n",
    "                stream = self._helper.skip_lines(file)\n",
    "                stream = self._helper.strip_newline(stream)\n",
    "                stream = self._helper.decode(stream)\n",
    "                yield from self._helper.return_path(stream, path=path)  # Returns 1 line at a time as List[str or bytes]\n",
    "```\n",
    "\n",
    "</blockquote>"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Class constructors vs. Functional form <a class=\"anchor\" id=\"11\"></a>\n",
    "\n",
    "DataPipes can be used via their class construction or functional form (a method call). The PyTorch team recommends using the funtional form, which is available to most but not all DataPipes. For example:\n",
    "\n",
    "Using *class constructors*:\n",
    "\n",
    "```python\n",
    "    new_dp = dp.iter.FileOpener([csv_file])\n",
    "    new_dp = dp.iter.CSVParser(new_dp, skip_lines=1)\n",
    "    # returns tuples like ('train/0/16585.png', '0')\n",
    "```\n",
    "\n",
    "Using *functional form*:\n",
    "\n",
    "```python\n",
    "    new_dp = dp.iter.FileOpener([csv_file])\n",
    "    new_dp = new_dp.parse_csv(skip_lines=1)\n",
    "    # returns tuples like ('train/0/16585.png', '0')\n",
    "```"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "A pipeline can be assembled with a mix of using class constructors and their functional forms, for example:\n",
    "\n",
    "```python\n",
    "    FOLDER = 'path/2/csv/folder'\n",
    "    datapipe = dp.iter.FileLister([FOLDER]).filter(filter_fn=lambda filename: filename.endswith('.csv'))\n",
    "    datapipe = dp.iter.FileOpener(datapipe, mode='rt')\n",
    "    datapipe = datapipe.parse_csv(delimiter=',')\n",
    "    N_ROWS = 10000  # total number of rows of data\n",
    "    train, valid = datapipe.random_split(total_length=N_ROWS, weights={\"train\": 0.5, \"valid\": 0.5}, seed=0)\n",
    "\n",
    "    for x in train:  # Iterating through the training dataset\n",
    "        pass\n",
    "\n",
    "    for y in valid:  # Iterating through the validation dataset\n",
    "        pass\n",
    "```"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### IterDataPipes vs. MapDataPipes <a class=\"anchor\" id=\"12\"></a>\n",
    "\n",
    "\n",
    "#### Map-style DataPipes\n",
    "\n",
    "From the [Map-style DataPipes Docs](https://pytorch.org/data/main/torchdata.datapipes.map.html):\n",
    "\n",
    "<blockquote>\n",
    "\n",
    "A Map-style DataPipe is one that implements the `__getitem__()` and `__len__()` protocols, and represents a map from (possibly non-integral) indices/keys to data samples. This is a close equivalent of `Dataset` from the PyTorch core library.\n",
    "\n",
    "For example, when accessed with `mapdatapipe[idx]`, could read the idx-th image and its corresponding label from a folder on the disk.\n",
    "\n",
    "All datasets that represent a map from keys to data samples should subclass `torchdata.datapipes.map.MapDataPipe` and overwrite `__getitem__()` and (optionally) `__len__()`.\n",
    "\n",
    "</blockquote>\n",
    "\n",
    "\n",
    "\n",
    "#### Iterable-style DataPipes\n",
    "\n",
    "From the [Iterable-style DataPipes Docs](https://pytorch.org/data/main/torchdata.datapipes.iter.html):\n",
    "\n",
    "<blockquote>\n",
    "\n",
    "An iterable-style dataset is an instance of a subclass of IterableDataset that implements the `__iter__()` protocol, and represents an iterable over data samples. This type of datasets is particularly suitable for cases where random reads are expensive or even improbable, and where the batch size depends on the fetched data.\n",
    "\n",
    "For example, such a dataset, when called `iter(iterdatapipe)`, could return a stream of data reading from a database, a remote server, or even logs generated in real time.\n",
    "\n",
    "This is an updated version of `IterableDataset` in torch.\n",
    "\n",
    "All DataPipes that represent an iterable of data samples should subclass `torchdata.datapipes.iter.IterDataPipe`.\n",
    "\n",
    "</blockquote>\n",
    "\n",
    "\n",
    "\n",
    "Full lists of the available DataPipes can be found in the above linked doc pages.\n",
    "\n",
    "#### Conversion\n",
    "\n",
    "An `IterDataPipe` can be converted to a `MapDataPipe` using either `IterToMapConverter` or `.to_map_datapipe()`. \n",
    "\n",
    "Currently TorchData has placed more emphasis and effort on developing `IterDataPipe`s, so there are many more available compared to `MapDataPipe`s."
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----\n",
    "\n",
    "## Tabular Data Examples [[5]](https://pytorch.org/data/0.5/tutorial.html)[[6]](https://github.com/pytorch/data#frequently-asked-questions-faq) <a class=\"anchor\" id=\"13\"></a>\n",
    "\n",
    "The below code examples are borrowed from the official TorchData tutorial [5] of creating a DataPipe for tabular CSV data using TorchData's functional form.\n",
    "\n",
    "Lets first use a helper function to generate some CSV files with random data and labels."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "sample_data0.csv  sample_data1.csv  sample_data2.csv\n",
      "label,c0,c1,c2\n",
      "4,0.14753037391393564,0.5640690419198052,0.8431982458786857\n",
      "5,0.202883724580455,0.24674133527166886,0.21224273558917006\n"
     ]
    }
   ],
   "source": [
    "import csv\n",
    "import random\n",
    "\n",
    "!mkdir -p data/csv\n",
    "\n",
    "def generate_csv(file_label, num_rows: int = 5000, num_features: int = 20) -> None:\n",
    "    fieldnames = ['label'] + [f'c{i}' for i in range(num_features)]\n",
    "    writer = csv.DictWriter(open(f\"data/csv/sample_data{file_label}.csv\", \"w\", newline=''), fieldnames=fieldnames)\n",
    "    writer.writeheader()\n",
    "    for i in range(num_rows):\n",
    "        row_data = {col: random.random() for col in fieldnames}\n",
    "        row_data['label'] = random.randint(0, 9)\n",
    "        writer.writerow(row_data)\n",
    "\n",
    "# Generate 3 csv files with 10 rows and 3 features\n",
    "num_files_to_generate = 3\n",
    "for i in range(num_files_to_generate):\n",
    "    generate_csv(file_label=i, num_rows=10, num_features=3)\n",
    "\n",
    "# List the generated csv files\n",
    "!ls data/csv\n",
    "\n",
    "# Print the first 3 lines of the first csv file\n",
    "!head -n3 data/csv/sample_data0.csv"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Once we have our CSV data, we can build a `DataPipe` to read and parse it. \n",
    "\n",
    "It is preferable to define functions that are passed to DataPipes rather than using `lambda` functions, because `lambda` functions cannot be serialized with `pickle`. If you install the package `dill` then lambda functions will be serializable. However if you don't have `dill` installed, you'll receive a warning about unserializable functions. The use of unserializable functions will prevent multiprocessing from being used.\n",
    "\n",
    "Also note that you may wrap a function with `functools.partial` or use the function `operator.itemgetter` to simplify `.map(function)` calls."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import torchdata.datapipes as dp\n",
    "\n",
    "\n",
    "def filter_for_data(filename):\n",
    "    return \"sample_data\" in filename and filename.endswith(\".csv\")\n",
    "\n",
    "def row_processer(row):\n",
    "    return {\"label\": np.array(row[0], np.int32), \"data\": np.array(row[1:], dtype=np.float64)}\n",
    "\n",
    "def build_datapipes(root_dir=\".\"):\n",
    "    datapipe = dp.iter.FileLister(root_dir)\n",
    "    datapipe = datapipe.filter(filter_fn=filter_for_data)\n",
    "    datapipe = datapipe.open_files(mode='rt')\n",
    "    datapipe = datapipe.parse_csv(delimiter=\",\", skip_lines=1)\n",
    "    # Shuffle will happen as long as you do NOT set `shuffle=False` later in the DataLoader\n",
    "    datapipe = datapipe.shuffle()\n",
    "    # Make sure distributed sharding is mutually exclusive and collectively exhaustive\n",
    "    datapipe = datapipe.sharding_filter()\n",
    "    datapipe = datapipe.map(row_processer)\n",
    "    return datapipe"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Using a DataPipe with DataLoader <a class=\"anchor\" id=\"14\"></a>\n",
    "\n",
    "Use a `DataPipe` with `DataLoader` by passing `dataset=datapipe` as an argument. Set shuffling with `DataPipe` and batching and multiprocessing with `DataLoader`. For more information, refer to the [DataLoader documentation](https://pytorch.org/docs/stable/data.html#single-and-multi-process-data-loading). If using `Batcher` with `DataLoader` with `batch_size > 1`, samples will be batched multiple times, so don't apply batching both in the DataPipe and DataLoader."
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "\n",
    "#### Multiprocessing important caveats <a class=\"anchor\" id=\"15\"></a>\n",
    "\n",
    "<blockquote>\n",
    "\n",
    "- Multi-process data loading is still handled by the `DataLoader`, see the [DataLoader documentation for more details](https://pytorch.org/docs/stable/data.html#single-and-multi-process-data-loading). As of PyTorch version >= 1.12.0 (TorchData version >= 0.4.0), data sharding is automatically done for `DataPipes` within the `DataLoader` as long as a `ShardingFilter` (`datapipe.sharding_filter`) DataPipe exists in your pipeline. Sharding filter is a necessary workaround to avoid data duplication when we use more than 1 worker.\n",
    "- Place `ShardingFilter` (`datapipe.sharding_filter`) as early as possible in the pipeline, especially before expensive operations such as decoding, in order to avoid repeating these expensive operations across worker/distributed processes.\n",
    "- For the data source that needs to be sharded, it is crucial to add Shuffler before ShardingFilter to ensure data are globally shuffled before splitted into shards. Otherwise, each worker process would always process the same shard of data for all epochs. And, it means each batch would only consist of data from the same shard, which leads to low accuracy during training. However, it doesn’t apply to the data source that has already been sharded for each multi-/distributed process, since ShardingFilter is no longer required to be presented in the pipeline.\n",
    "- There may be cases where placing Shuffler earlier in the pipeline lead to worse performance, because some operations (e.g. decompression) are faster with sequential reading. In those cases, we recommend decompressing the files prior to shuffling (potentially prior to any data loading).\n",
    "\n",
    "</blockquote>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.utils.data import DataLoader\n",
    "\n",
    "datapipe = build_datapipes(root_dir='./data/csv')\n",
    "dl = DataLoader(dataset=datapipe, batch_size=5, num_workers=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Labels batch shape: torch.Size([5])\n",
      "Feature batch shape: torch.Size([5, 3])\n",
      "labels = tensor([4, 5, 1, 5, 8], dtype=torch.int32)\n",
      "features = tensor([[0.0331, 0.9106, 0.8185],\n",
      "        [0.3699, 0.7761, 0.8073],\n",
      "        [0.8772, 0.3190, 0.1535],\n",
      "        [0.3282, 0.5670, 0.4941],\n",
      "        [0.3580, 0.2634, 0.6832]], dtype=torch.float64)\n",
      "n_sample = 6\n"
     ]
    }
   ],
   "source": [
    "batch = next(iter(dl))\n",
    "labels, features = batch['label'], batch['data']\n",
    "\n",
    "print(f\"Labels batch shape: {labels.size()}\")\n",
    "print(f\"Feature batch shape: {features.size()}\")\n",
    "print(f\"{labels = }\\n{features = }\")\n",
    "\n",
    "n_sample = 0\n",
    "for row in iter(dl):\n",
    "    n_sample += 1\n",
    "print(f\"{n_sample = }\")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### DataLoader vs [DataLoader2](https://pytorch.org/data/beta/dataloader2.html#id1) [[7]](https://pytorch.org/data/beta/dataloader2.html#module-torchdata.dataloader2) <a class=\"anchor\" id=\"16\"></a>\n",
    "\n",
    "#### Using a DataPipe with DataLoader2\n",
    "\n",
    "TorchData's light-weight `torchdata.dataloader2.DataLoader2` no longer includes batching or shuffling options to avoid confusion. We'll do shuffling and batching by the DataPipe, and multiprocessing via `DataLoader2`'s `reading_service`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torchdata.dataloader2 import (DataLoader2,\n",
    "                                   PrototypeMultiProcessingReadingService)\n",
    "\n",
    "# We'll add batching and collating (with default collate_fn) to the datapipe\n",
    "datapipe = build_datapipes(root_dir='./data/csv').batch(batch_size=5).collate()\n",
    "\n",
    "# PrototypeMultiProcessingReadingService is subject to change, check the notes on this for more details:\n",
    "# https://github.com/pytorch/data/blob/main/torchdata/dataloader2/reading_service.py#L124\n",
    "reading_service = PrototypeMultiProcessingReadingService(num_workers=2)\n",
    "\n",
    "dl = DataLoader2(datapipe=datapipe, reading_service=reading_service)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Labels batch shape: torch.Size([5])\n",
      "Feature batch shape: torch.Size([5, 3])\n",
      "labels = tensor([0, 2, 3, 2, 3], dtype=torch.int32)\n",
      "features = tensor([[0.7787, 0.7525, 0.4740],\n",
      "        [0.5130, 0.3275, 0.3772],\n",
      "        [0.7539, 0.4334, 0.4641],\n",
      "        [0.8932, 0.7525, 0.6826],\n",
      "        [0.3357, 0.8314, 0.6937]], dtype=torch.float64)\n",
      "n_sample = 6\n"
     ]
    }
   ],
   "source": [
    "batch = next(iter(dl))\n",
    "labels, features = batch['label'], batch['data']\n",
    "\n",
    "print(f\"Labels batch shape: {labels.size()}\")\n",
    "print(f\"Feature batch shape: {features.size()}\")\n",
    "print(f\"{labels = }\\n{features = }\")\n",
    "\n",
    "n_sample = 0\n",
    "for row in iter(dl):\n",
    "    n_sample += 1\n",
    "print(f\"{n_sample = }\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ShufflerIterDataPipe instance doesn't have valid length\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "6"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Notice iter DataPipe has no length by default\n",
    "try:\n",
    "    len(datapipe)\n",
    "except Exception as e:\n",
    "    print(e)\n",
    "\n",
    "# But it can be set\n",
    "datapipe = datapipe.set_length(len(list(datapipe)))\n",
    "len(datapipe)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Splitting data <a class=\"anchor\" id=\"17\"></a>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(15, 15)"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import numpy as np\n",
    "import torchdata.datapipes as dp\n",
    "\n",
    "\n",
    "def filter_for_data(filename):\n",
    "    return \"sample_data\" in filename and filename.endswith(\".csv\")\n",
    "\n",
    "def row_processer(row):\n",
    "    return {\"label\": np.array(row[0], np.int32), \"data\": np.array(row[1:], dtype=np.float64)}\n",
    "\n",
    "def build_datapipes(N_ROWS, batch_size=5, root_dir=\".\"):\n",
    "    datapipe = dp.iter.FileLister(root_dir)\n",
    "    datapipe = datapipe.filter(filter_fn=filter_for_data)\n",
    "    datapipe = datapipe.open_files(mode='rt')\n",
    "    datapipe = datapipe.parse_csv(delimiter=\",\", skip_lines=1)\n",
    "    # Shuffle will happen as long as you do NOT set `shuffle=False` later in the DataLoader\n",
    "    datapipe = datapipe.shuffle()\n",
    "    datapipe = datapipe.map(row_processer)\n",
    "    datapipe = datapipe.sharding_filter()\n",
    "    datapipe = datapipe.batch(batch_size)\n",
    "    datapipe = datapipe.collate()\n",
    "    train, valid = datapipe.random_split(total_length=N_ROWS, weights={\"train\": 0.5, \"valid\": 0.5}, seed=0)\n",
    "    return train, valid\n",
    "\n",
    "N_ROWS = 30\n",
    "train, valid = build_datapipes(N_ROWS, root_dir='./data/csv')\n",
    "\n",
    "for batch in train:\n",
    "    pass\n",
    "for batch in valid:\n",
    "    pass\n",
    "\n",
    "# Since random_split required total_length, now we have __len__ for datapipes\n",
    "len(train), len(valid)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Computing the number of samples [[8]](https://github.com/pytorch/vision/tree/main/torchvision/prototype/datasets/_builtin#how-do-i-compute-the-number-of-samples) <a class=\"anchor\" id=\"18\"></a>\n",
    "\n",
    "One downside of iterable-style datasets is that there's no way to check their length other than to iterate over the dataset and count the number of samples. It's a good idea to disable all unncecessary operations when doing this to speed up the process. \n",
    "\n",
    "```python\n",
    "    import itertools\n",
    "    from torchvision.prototype import datasets\n",
    "\n",
    "\n",
    "    def combinations_grid(**kwargs):\n",
    "        return [dict(zip(kwargs.keys(), values)) for values in itertools.product(*kwargs.values())]\n",
    "\n",
    "\n",
    "    # If you have implemented the mock data function for the dataset tests, you can simply copy-paste from there\n",
    "    configs = combinations_grid(split=(\"train\", \"test\"), foo=(\"bar\", \"baz\"))\n",
    "\n",
    "    for config in configs:\n",
    "        dataset = datasets.load(\"my-dataset\", **config)\n",
    "\n",
    "        num_samples = 0\n",
    "        for _ in dataset:\n",
    "            num_samples += 1\n",
    "\n",
    "        print(\", \".join(f\"{key}={value}\" for key, value in config.items()), num_samples)\n",
    "```"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### TorchArrow [[9]](https://github.com/pytorch/torcharrow#torcharrow-a-data-processing-library-for-pytorch) <a class=\"anchor\" id=\"19\"></a>\n",
    "\n",
    "<blockquote>\n",
    "\n",
    "TorchArrow is a `torch.Tensor`-like Python DataFrame library for data preprocessing in PyTorch models, with two high-level features:\n",
    "\n",
    "- DataFrame library (like Pandas) with strong GPU or other hardware acceleration (under development) and PyTorch ecosystem integration.\n",
    "- Columnar memory layout based on Apache Arrow with strong variable-width and nested data support (such as string, list, map) and Arrow ecosystem integration.\n",
    "\n",
    "TorchArrow is currently integrated with TorchData in two ways:\n",
    "- `DataFrameMaker`: Takes rows of data, batches a number of them together and creates TorchArrow DataFrames (functional name: `dataframe`).\n",
    "- `ParquetDataFrameLoader`: Takes in paths to Parquet files and return a TorchArrow DataFrame for each row group within a Parquet file (functional name: `load_parquet_as_df`).\n",
    "\n",
    "</blockquote>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torcharrow as ta\n",
    "import torcharrow.dtypes as dt\n",
    "from torchdata.datapipes.iter import (DataFrameMaker, FileLister,\n",
    "                                      IterableWrapper, ParquetDataFrameLoader)\n",
    "\n",
    "device = 'cpu'"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### DataFrameMaker"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create example DataPipe\n",
    "source_data = [(i,) for i in range(10)]\n",
    "source_dp = IterableWrapper(source_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'torcharrow.velox_rt.dataframe_cpu.DataFrameCpu'>\n",
      "10\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "  index    Values\n",
       "-------  --------\n",
       "      0         0\n",
       "      1         1\n",
       "dtype: Struct([Field('Values', int32)]), count: 2, null_count: 0"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Take rows of data and convert to single TorchArrow DataFrame\n",
    "DTYPE = dt.Struct([dt.Field(\"Values\", dt.int32)])\n",
    "df_dp = source_dp.dataframe(dtype=DTYPE)\n",
    "df = list(df_dp)[0]\n",
    "\n",
    "print(type(df))\n",
    "print(df.length)\n",
    "df.head(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2\n",
      "5\n",
      "  index    Values\n",
      "-------  --------\n",
      "      0         0\n",
      "      1         1\n",
      "dtype: Struct([Field('Values', int32)]), count: 2, null_count: 0\n",
      "5\n",
      "  index    Values\n",
      "-------  --------\n",
      "      0         5\n",
      "      1         6\n",
      "dtype: Struct([Field('Values', int32)]), count: 2, null_count: 0\n"
     ]
    }
   ],
   "source": [
    "# Take DataPipe and convert into multiple TorchArrow DataFrames, based on size argument \n",
    "df_dp = DataFrameMaker(source_dp, dataframe_size=5, dtype=DTYPE, device=device, dtype_generator=None, columns=None)\n",
    "dfs = list(df_dp)\n",
    "\n",
    "print(len(dfs))\n",
    "for df in dfs:\n",
    "    print(df.length)\n",
    "    print(df.head(2))"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### ParquetDataFrameLoader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create TorchArrow DataFrames\n",
    "DTYPE = dt.Struct([dt.Field(\"Values\", dt.int32)])\n",
    "df1 = ta.dataframe([(i,) for i in range(10)], dtype=DTYPE)\n",
    "df2 = ta.dataframe([(i,) for i in range(100)], dtype=DTYPE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "df0.parquet  df1.parquet  merged.parquet\n"
     ]
    }
   ],
   "source": [
    "# Write them as parquet files\n",
    "import os\n",
    "import pyarrow\n",
    "import pyarrow.parquet as parquet\n",
    "\n",
    "!mkdir -p data/parquet\n",
    "dir_path = \"data/parquet\"\n",
    "\n",
    "for i, df in enumerate([df1, df2]):\n",
    "\n",
    "    # Write df as parquet\n",
    "    table = df.to_arrow()\n",
    "    parquet.write_table(table, os.path.join(dir_path, f\"df{i}.parquet\"))\n",
    "\n",
    "    # write multiple dfs as parquet\n",
    "    dfs = [df1, df2]\n",
    "    tables = [df.to_arrow() for df in dfs]\n",
    "    merged_table = pyarrow.concat_tables(tables)\n",
    "    parquet.write_table(merged_table, os.path.join(dir_path, \"merged.parquet\"))\n",
    "\n",
    "!ls data/parquet"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'torcharrow.velox_rt.dataframe_cpu.DataFrameCpu'> 10\n",
      "<class 'torcharrow.velox_rt.dataframe_cpu.DataFrameCpu'> 100\n"
     ]
    }
   ],
   "source": [
    "# Read from Parquet files and output TorchArrow DataFrames\n",
    "DTYPE = dt.Struct([dt.Field(\"Values\", dt.int32)])\n",
    "source_dp = FileLister(dir_path, masks=\"df*.parquet\")\n",
    "parquet_df_dp = ParquetDataFrameLoader(source_dp, dtype=DTYPE)\n",
    "\n",
    "for df in parquet_df_dp:\n",
    "    print(type(df), df.length)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'torcharrow.velox_rt.dataframe_cpu.DataFrameCpu'> 110\n"
     ]
    }
   ],
   "source": [
    "# Read from a Parquet file that was a merged DataFrame\n",
    "merged_source_dp = FileLister(dir_path, masks=\"merged.parquet\")\n",
    "merged_parquet_df_dp = ParquetDataFrameLoader(merged_source_dp, dtype=DTYPE, device=device, use_threads=True, columns=None)\n",
    "\n",
    "for df in merged_parquet_df_dp:\n",
    "    print(type(df), df.length)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Vision Data Examples <a class=\"anchor\" id=\"20\"></a>\n",
    "\n",
    "### Torchvision <a class=\"anchor\" id=\"21\"></a>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['caltech101',\n",
       " 'caltech256',\n",
       " 'celeba',\n",
       " 'cifar10',\n",
       " 'cifar100',\n",
       " 'clevr',\n",
       " 'coco',\n",
       " 'country211',\n",
       " 'cub200',\n",
       " 'dtd',\n",
       " 'emnist',\n",
       " 'eurosat',\n",
       " 'fashionmnist',\n",
       " 'fer2013',\n",
       " 'food101',\n",
       " 'gtsrb',\n",
       " 'imagenet',\n",
       " 'kmnist',\n",
       " 'mnist',\n",
       " 'oxford-iiit-pet',\n",
       " 'pcam',\n",
       " 'qmnist',\n",
       " 'sbd',\n",
       " 'semeion',\n",
       " 'stanford-cars',\n",
       " 'svhn',\n",
       " 'usps',\n",
       " 'voc']"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from torchvision.prototype import datasets\n",
    "datasets.list_datasets()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'categories': ['airplane',\n",
       "  'automobile',\n",
       "  'bird',\n",
       "  'cat',\n",
       "  'deer',\n",
       "  'dog',\n",
       "  'frog',\n",
       "  'horse',\n",
       "  'ship',\n",
       "  'truck']}"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "datasets.info(\"cifar10\")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Most built-in datasets can be downloaded automatically when calling `dataset.load()`, some cannot due to license restrictions (for example ImageNet). In this case you will receive a `RuntimeError` that will provide instructions for downloading."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'image': Image([[[178, 178, 178,  ..., 170, 168, 165],\n",
      "        [180, 179, 180,  ..., 173, 171, 168],\n",
      "        [177, 177, 178,  ..., 171, 169, 167],\n",
      "        ...,\n",
      "        [112, 113, 114,  ..., 100,  98, 101],\n",
      "        [112, 112, 113,  ..., 102, 102, 102],\n",
      "        [103, 100, 103,  ...,  92,  93,  91]],\n",
      "\n",
      "       [[176, 176, 176,  ..., 168, 166, 163],\n",
      "        [178, 177, 178,  ..., 171, 169, 166],\n",
      "        [175, 175, 176,  ..., 169, 167, 165],\n",
      "        ...,\n",
      "        [107, 109, 110,  ...,  97,  94,  95],\n",
      "        [102, 103, 103,  ...,  95,  93,  92],\n",
      "        [ 96,  93,  95,  ...,  84,  86,  84]],\n",
      "\n",
      "       [[189, 189, 189,  ..., 180, 177, 174],\n",
      "        [191, 190, 191,  ..., 182, 180, 177],\n",
      "        [188, 188, 189,  ..., 180, 178, 176],\n",
      "        ...,\n",
      "        [107, 108, 110,  ...,  94,  93,  95],\n",
      "        [101, 102, 103,  ...,  93,  91,  91],\n",
      "        [ 92,  90,  94,  ...,  80,  80,  77]]], dtype=torch.uint8, ), 'label': Label(0)}\n"
     ]
    }
   ],
   "source": [
    "from torchdata.datapipes.iter import IterDataPipe\n",
    "from torchvision.prototype import datasets\n",
    "\n",
    "dataset = datasets.load(\"cifar10\")\n",
    "for sample in dataset:\n",
    "    print(sample)  # this is the content of an item in datapipe returned by _datapipe()\n",
    "    break\n",
    "\n",
    "# Notice that dataset is a subclass of IterDataPipe\n",
    "assert issubclass(type(dataset), IterDataPipe)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cifar10\n"
     ]
    }
   ],
   "source": [
    "# By default (if `root` is not provided), downloads will be saved in ~/.cache/torch/datasets/vision/\n",
    "!ls ~/.cache/torch/datasets/vision/"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Image Folder [[10]](https://github.com/pytorch/data/blob/main/examples/vision/imagefolder.py) <a class=\"anchor\" id=\"22\"></a>\n",
    "\n",
    "<blockquote>\n",
    "\n",
    "Torchvision has the familiar and convenient `torchvision.datasets.ImageFolder` class, which is a generic dataloader that inherits from `DatasetFolder`. We can recreate the same functionality, but instead using DataPipes. The below DataPipes implementation of ImageFolder constructs and executes a graph of DataPipes (aka DataPipeline):\n",
    "\n",
    "```text\n",
    "FileLister -> ObtainCategories\n",
    "                      ⬇\n",
    "FileLister -> AttributeCategories -> LoadAndDecodeImages (using `map`) -> ApplyTorchVisionTransforms (using `map`)\n",
    "```\n",
    "</blockquote>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "A    data/vision/imagefolder/cat\n",
      "A    data/vision/imagefolder/cat/1.jpg\n",
      "A    data/vision/imagefolder/cat/2.jpg\n",
      "A    data/vision/imagefolder/cat/3.jpg\n",
      "A    data/vision/imagefolder/dog\n",
      "A    data/vision/imagefolder/dog/1.jpg\n",
      "A    data/vision/imagefolder/dog/2.jpg\n",
      "A    data/vision/imagefolder/dog/3.jpg\n",
      "Checked out revision 1878.\n",
      "data/vision/imagefolder\n",
      "├── cat\n",
      "│   ├── 1.jpg\n",
      "│   ├── 2.jpg\n",
      "│   └── 3.jpg\n",
      "└── dog\n",
      "    ├── 1.jpg\n",
      "    ├── 2.jpg\n",
      "    └── 3.jpg\n",
      "\n",
      "2 directories, 6 files\n"
     ]
    }
   ],
   "source": [
    "# Download fake data for this example from GitHub\n",
    "!mkdir -p data/vision\n",
    "!svn checkout https://github.com/pytorch/data/trunk/examples/vision/fakedata/imagefolder data/vision/imagefolder\n",
    "!rm -rf data/vision/imagefolder/.svn\n",
    "\n",
    "# Check out the downloaded directory\n",
    "!tree data/vision/imagefolder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {},
   "outputs": [],
   "source": [
    "import http.server\n",
    "import os\n",
    "import re\n",
    "import threading\n",
    "\n",
    "import torchvision.datasets as datasets\n",
    "import torchvision.datasets.folder\n",
    "import torchvision.transforms as transforms\n",
    "from PIL import Image\n",
    "from torch.utils.data import DataLoader\n",
    "from torchdata.datapipes.iter import FileLister, HttpReader, IterDataPipe\n",
    "\n",
    "\n",
    "IMAGES_ROOT = \"data/vision/imagefolder\"\n",
    "\n",
    "USE_FORK_DATAPIPE = False\n",
    "NUM_WORKERS = 8\n",
    "BATCH_SIZE = 32\n",
    "\n",
    "data_transform = transforms.Compose(\n",
    "    [\n",
    "        transforms.RandomResizedCrop(224),\n",
    "        transforms.RandomHorizontalFlip(),\n",
    "        transforms.ToTensor(),\n",
    "        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),\n",
    "    ]\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_category_name(path):\n",
    "    rel_path = os.path.relpath(path, start=IMAGES_ROOT)\n",
    "    elements = rel_path.split(os.sep)\n",
    "    return elements[0]\n",
    "\n",
    "\n",
    "class ObtainCategories(IterDataPipe):\n",
    "    \"\"\" Creates a DataPipe that yields a dictionary of categories to IDs. \"\"\"\n",
    "    def __init__(self, source_dp, parse_category_fn=get_category_name) -> None:\n",
    "        self.source_dp = source_dp\n",
    "        self.parse_category_fn = parse_category_fn\n",
    "\n",
    "    def __iter__(self):\n",
    "        categories = set()\n",
    "        for path in self.source_dp:\n",
    "            categories.add(self.parse_category_fn(path))\n",
    "        cat_to_id = {name: i for i, name in enumerate(sorted(categories))}\n",
    "        yield cat_to_id\n",
    "\n",
    "\n",
    "class AttributeCategories(IterDataPipe):\n",
    "    \"\"\" Creates a DataPipe that yields a tuple of (data, category) \"\"\"\n",
    "    def __init__(self, listfiles_dp, categories_dp, parse_category_fn=get_category_name) -> None:\n",
    "        self.listfiles_dp = listfiles_dp\n",
    "        self.categories_dp = categories_dp\n",
    "        self.parse_category_fn = parse_category_fn\n",
    "\n",
    "    def __iter__(self):\n",
    "        for categories in self.categories_dp:\n",
    "            cat_to_dp = categories\n",
    "        for data in self.listfiles_dp:\n",
    "            if isinstance(data, tuple):\n",
    "                category = cat_to_dp[self.parse_category_fn(data[0])]\n",
    "                yield data + (category,)\n",
    "            else:\n",
    "                category = cat_to_dp[self.parse_category_fn(data)]\n",
    "                yield (data, category)\n",
    "\n",
    "\n",
    "def MyImageFolder(root=IMAGES_ROOT, transform=None):\n",
    "    if not USE_FORK_DATAPIPE:\n",
    "        # Yes, we had to scan files twice. Alternativelly it is possible to use\n",
    "        # `fork` DataPipe, but it will require buffer equal to the size of all\n",
    "        # full file names\n",
    "        # TODO(125): Make sure that `fork` complains when buffer becomes too large\n",
    "        list_files_0 = FileLister(root=root, recursive=True)\n",
    "        list_files_1 = FileLister(root=root, recursive=True).sharding_filter()\n",
    "    else:\n",
    "        list_files_0, list_files_1 = FileLister(root=root, recursive=True).fork(2)\n",
    "        list_files_1 = list_files_1.sharding_filter()\n",
    "\n",
    "    categories = ObtainCategories(list_files_0)\n",
    "    with_categories = AttributeCategories(list_files_1, categories)\n",
    "    using_default_loader = with_categories.map(lambda x: (torchvision.datasets.folder.default_loader(x[0]), x[1]))\n",
    "    transformed = using_default_loader.map(lambda x: (transform(x[0]), x[1]))\n",
    "    return transformed"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([1, 3, 224, 224]) tensor([0])\n"
     ]
    }
   ],
   "source": [
    "dataset = MyImageFolder(root=IMAGES_ROOT, transform=data_transform)\n",
    "\n",
    "dl = DataLoader(\n",
    "    dataset,\n",
    "    batch_size=BATCH_SIZE,\n",
    "    shuffle=False,\n",
    "    num_workers=NUM_WORKERS,\n",
    ")\n",
    "\n",
    "items = list(dl)\n",
    "assert len(items) == 6\n",
    "\n",
    "# Check first image size and class ID\n",
    "print(items[0][0].shape, items[0][1])"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Visualizing DataPipes <a class=\"anchor\" id=\"23\"></a>\n",
    "\n",
    "We can also use a built-in graphviz utility to visualize our DataPipes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/svg+xml": "<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"no\"?>\n<!DOCTYPE svg PUBLIC \"-//W3C//DTD SVG 1.1//EN\"\n \"http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd\">\n<!-- Generated by graphviz version 2.30.1 (20180828.1746)\n -->\n<!-- Title: %3 Pages: 1 -->\n<svg width=\"242pt\" height=\"308pt\"\n viewBox=\"0.00 0.00 242.00 308.00\" xmlns=\"http://www.w3.org/2000/svg\" xmlns:xlink=\"http://www.w3.org/1999/xlink\">\n<g id=\"graph0\" class=\"graph\" transform=\"scale(1 1) rotate(0) translate(4 304)\">\n<title>%3</title>\n<polygon fill=\"white\" stroke=\"white\" points=\"-4,5 -4,-304 239,-304 239,5 -4,5\"/>\n<!-- Mapper&#45;8793610889698 -->\n<g id=\"node1\" class=\"node\"><title>Mapper&#45;8793610889698</title>\n<polygon fill=\"#caff70\" stroke=\"black\" points=\"145,-19.5 91,-19.5 91,-0.5 145,-0.5 145,-19.5\"/>\n<text text-anchor=\"middle\" x=\"118\" y=\"-7.5\" font-family=\"monospace\" font-size=\"10.00\">Mapper</text>\n</g>\n<!-- FileLister&#45;8793610889674 -->\n<g id=\"node2\" class=\"node\"><title>FileLister&#45;8793610889674</title>\n<polygon fill=\"lightgrey\" stroke=\"black\" points=\"94.25,-243.5 17.75,-243.5 17.75,-224.5 94.25,-224.5 94.25,-243.5\"/>\n<text text-anchor=\"middle\" x=\"56\" y=\"-231.5\" font-family=\"monospace\" font-size=\"10.00\">FileLister</text>\n</g>\n<!-- ObtainCategories&#45;8793610889683 -->\n<g id=\"node3\" class=\"node\"><title>ObtainCategories&#45;8793610889683</title>\n<polygon fill=\"lightgrey\" stroke=\"black\" points=\"112.25,-187.5 -0.25,-187.5 -0.25,-168.5 112.25,-168.5 112.25,-187.5\"/>\n<text text-anchor=\"middle\" x=\"56\" y=\"-175.5\" font-family=\"monospace\" font-size=\"10.00\">ObtainCategories</text>\n</g>\n<!-- FileLister&#45;8793610889674&#45;&gt;ObtainCategories&#45;8793610889683 -->\n<g id=\"edge1\" class=\"edge\"><title>FileLister&#45;8793610889674&#45;&gt;ObtainCategories&#45;8793610889683</title>\n<path fill=\"none\" stroke=\"black\" d=\"M56,-224.083C56,-217.006 56,-206.861 56,-197.986\"/>\n<polygon fill=\"black\" stroke=\"black\" points=\"59.5001,-197.751 56,-187.751 52.5001,-197.751 59.5001,-197.751\"/>\n</g>\n<!-- AttributeCategories&#45;8793610889689 -->\n<g id=\"node5\" class=\"node\"><title>AttributeCategories&#45;8793610889689</title>\n<polygon fill=\"lightgrey\" stroke=\"black\" points=\"183.25,-131.5 52.75,-131.5 52.75,-112.5 183.25,-112.5 183.25,-131.5\"/>\n<text text-anchor=\"middle\" x=\"118\" y=\"-119.5\" font-family=\"monospace\" font-size=\"10.00\">AttributeCategories</text>\n</g>\n<!-- ObtainCategories&#45;8793610889683&#45;&gt;AttributeCategories&#45;8793610889689 -->\n<g id=\"edge5\" class=\"edge\"><title>ObtainCategories&#45;8793610889683&#45;&gt;AttributeCategories&#45;8793610889689</title>\n<path fill=\"none\" stroke=\"black\" d=\"M65.9614,-168.324C75.1266,-160.341 88.9372,-148.313 100.028,-138.653\"/>\n<polygon fill=\"black\" stroke=\"black\" points=\"102.615,-141.041 107.857,-131.834 98.0178,-135.762 102.615,-141.041\"/>\n</g>\n<!-- ShardingFilter&#45;8793610889677 -->\n<g id=\"node4\" class=\"node\"><title>ShardingFilter&#45;8793610889677</title>\n<polygon fill=\"lightgrey\" stroke=\"black\" points=\"231.25,-187.5 130.75,-187.5 130.75,-168.5 231.25,-168.5 231.25,-187.5\"/>\n<text text-anchor=\"middle\" x=\"181\" y=\"-175.5\" font-family=\"monospace\" font-size=\"10.00\">ShardingFilter</text>\n</g>\n<!-- ShardingFilter&#45;8793610889677&#45;&gt;AttributeCategories&#45;8793610889689 -->\n<g id=\"edge2\" class=\"edge\"><title>ShardingFilter&#45;8793610889677&#45;&gt;AttributeCategories&#45;8793610889689</title>\n<path fill=\"none\" stroke=\"black\" d=\"M170.878,-168.324C161.474,-160.263 147.256,-148.077 135.931,-138.369\"/>\n<polygon fill=\"black\" stroke=\"black\" points=\"138.177,-135.684 128.306,-131.834 133.621,-140.999 138.177,-135.684\"/>\n</g>\n<!-- Mapper&#45;8793610889695 -->\n<g id=\"node8\" class=\"node\"><title>Mapper&#45;8793610889695</title>\n<polygon fill=\"lightgrey\" stroke=\"black\" points=\"145,-75.5 91,-75.5 91,-56.5 145,-56.5 145,-75.5\"/>\n<text text-anchor=\"middle\" x=\"118\" y=\"-63.5\" font-family=\"monospace\" font-size=\"10.00\">Mapper</text>\n</g>\n<!-- AttributeCategories&#45;8793610889689&#45;&gt;Mapper&#45;8793610889695 -->\n<g id=\"edge6\" class=\"edge\"><title>AttributeCategories&#45;8793610889689&#45;&gt;Mapper&#45;8793610889695</title>\n<path fill=\"none\" stroke=\"black\" d=\"M118,-112.083C118,-105.006 118,-94.8611 118,-85.9865\"/>\n<polygon fill=\"black\" stroke=\"black\" points=\"121.5,-85.751 118,-75.7511 114.5,-85.7511 121.5,-85.751\"/>\n</g>\n<!-- IterableWrapper&#45;8793610889581 -->\n<g id=\"node6\" class=\"node\"><title>IterableWrapper&#45;8793610889581</title>\n<polygon fill=\"lightblue\" stroke=\"black\" points=\"234.25,-299.5 127.75,-299.5 127.75,-280.5 234.25,-280.5 234.25,-299.5\"/>\n<text text-anchor=\"middle\" x=\"181\" y=\"-287.5\" font-family=\"monospace\" font-size=\"10.00\">IterableWrapper</text>\n</g>\n<!-- FileLister&#45;8793610889680 -->\n<g id=\"node7\" class=\"node\"><title>FileLister&#45;8793610889680</title>\n<polygon fill=\"lightgrey\" stroke=\"black\" points=\"219.25,-243.5 142.75,-243.5 142.75,-224.5 219.25,-224.5 219.25,-243.5\"/>\n<text text-anchor=\"middle\" x=\"181\" y=\"-231.5\" font-family=\"monospace\" font-size=\"10.00\">FileLister</text>\n</g>\n<!-- IterableWrapper&#45;8793610889581&#45;&gt;FileLister&#45;8793610889680 -->\n<g id=\"edge3\" class=\"edge\"><title>IterableWrapper&#45;8793610889581&#45;&gt;FileLister&#45;8793610889680</title>\n<path fill=\"none\" stroke=\"black\" d=\"M181,-280.083C181,-273.006 181,-262.861 181,-253.986\"/>\n<polygon fill=\"black\" stroke=\"black\" points=\"184.5,-253.751 181,-243.751 177.5,-253.751 184.5,-253.751\"/>\n</g>\n<!-- FileLister&#45;8793610889680&#45;&gt;ShardingFilter&#45;8793610889677 -->\n<g id=\"edge4\" class=\"edge\"><title>FileLister&#45;8793610889680&#45;&gt;ShardingFilter&#45;8793610889677</title>\n<path fill=\"none\" stroke=\"black\" d=\"M181,-224.083C181,-217.006 181,-206.861 181,-197.986\"/>\n<polygon fill=\"black\" stroke=\"black\" points=\"184.5,-197.751 181,-187.751 177.5,-197.751 184.5,-197.751\"/>\n</g>\n<!-- Mapper&#45;8793610889695&#45;&gt;Mapper&#45;8793610889698 -->\n<g id=\"edge8\" class=\"edge\"><title>Mapper&#45;8793610889695&#45;&gt;Mapper&#45;8793610889698</title>\n<path fill=\"none\" stroke=\"black\" d=\"M118,-56.083C118,-49.0059 118,-38.8611 118,-29.9865\"/>\n<polygon fill=\"black\" stroke=\"black\" points=\"121.5,-29.751 118,-19.7511 114.5,-29.7511 121.5,-29.751\"/>\n</g>\n<!-- IterableWrapper&#45;8793610889500 -->\n<g id=\"node9\" class=\"node\"><title>IterableWrapper&#45;8793610889500</title>\n<polygon fill=\"lightblue\" stroke=\"black\" points=\"109.25,-299.5 2.75,-299.5 2.75,-280.5 109.25,-280.5 109.25,-299.5\"/>\n<text text-anchor=\"middle\" x=\"56\" y=\"-287.5\" font-family=\"monospace\" font-size=\"10.00\">IterableWrapper</text>\n</g>\n<!-- IterableWrapper&#45;8793610889500&#45;&gt;FileLister&#45;8793610889674 -->\n<g id=\"edge7\" class=\"edge\"><title>IterableWrapper&#45;8793610889500&#45;&gt;FileLister&#45;8793610889674</title>\n<path fill=\"none\" stroke=\"black\" d=\"M56,-280.083C56,-273.006 56,-262.861 56,-253.986\"/>\n<polygon fill=\"black\" stroke=\"black\" points=\"59.5001,-253.751 56,-243.751 52.5001,-253.751 59.5001,-253.751\"/>\n</g>\n</g>\n</svg>\n",
      "text/plain": [
       "<graphviz.graphs.Digraph at 0x7ff6c0da9a30>"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from torchdata.datapipes.utils import to_graph\n",
    "\n",
    "G = to_graph(dataset)\n",
    "G"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### WebDataset <a class=\"anchor\" id=\"24\"></a>\n",
    "\n",
    "The [WebDataset Format](https://github.com/webdataset/webdataset#the-webdataset-format) is a relatively new format based on sharded tar files, and specifically designed for training with large datasets. WebDataset pipelines make it possibly to achieve higher I/O rates, among other benefits.\n",
    "\n",
    "TorchData supports WebDataset through their WebDataset DataPipe: [torchdata.datapipes.iter.WebDataset](https://pytorch.org/data/beta/generated/torchdata.datapipes.iter.WebDataset.html#torchdata.datapipes.iter.WebDataset). WebDataset is also supported by NVIDIA DALI.\n",
    "\n",
    "Note that SageMaker's [Fast File Mode](https://docs.aws.amazon.com/sagemaker/latest/dg/model-access-training-data.html) is also [more performant when using larger files](https://aws.amazon.com/blogs/machine-learning/choose-the-best-data-source-for-your-amazon-sagemaker-training-job/) like WebDataset Format's sharded tar files.\n",
    "\n",
    "To start, we'll first download the [Dog Breed Identification dataset](https://www.kaggle.com/competitions/dog-breed-identification/data) from Kaggle and convert it to the WebDataset Format. Note that in order to complete this step you'll need the Kaggle CLI installed with an API token copied to `~/.kaggle/kaggle.json`. You'll also need to [accept the competition's terms](https://www.kaggle.com/competitions/dog-breed-identification/rules)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading dog-breed-identification.zip to data/vision/dog-breed-identification\n",
      " 99%|███████████████████████████████████████▋| 686M/691M [00:09<00:00, 84.6MB/s]\n",
      "100%|████████████████████████████████████████| 691M/691M [00:09<00:00, 73.8MB/s]\n"
     ]
    }
   ],
   "source": [
    "# Download the dataset\n",
    "!kaggle competitions download -c dog-breed-identification -p data/vision/dog-breed-identification\n",
    "!cd data/vision/dog-breed-identification && unzip -q dog-breed-identification.zip"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Convert Images and CSV labels to WebDataset Format <a class=\"anchor\" id=\"25\"></a>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "nimages: 10222\n",
      "# writing data/vision/webdataset/train-000000.tar 0 0.0 GB 0\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e280e7699b0d4a81805a899f2e63a9d5",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/10222 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "# writing data/vision/webdataset/train-000001.tar 256 0.0 GB 256\n",
      "# writing data/vision/webdataset/train-000002.tar 256 0.0 GB 512\n",
      "# writing data/vision/webdataset/train-000003.tar 256 0.0 GB 768\n",
      "# writing data/vision/webdataset/train-000004.tar 256 0.0 GB 1024\n",
      "# writing data/vision/webdataset/train-000005.tar 256 0.0 GB 1280\n",
      "# writing data/vision/webdataset/train-000006.tar 256 0.0 GB 1536\n",
      "# writing data/vision/webdataset/train-000007.tar 256 0.0 GB 1792\n",
      "# writing data/vision/webdataset/train-000008.tar 256 0.0 GB 2048\n",
      "# writing data/vision/webdataset/train-000009.tar 256 0.0 GB 2304\n",
      "# writing data/vision/webdataset/train-000010.tar 256 0.0 GB 2560\n",
      "# writing data/vision/webdataset/train-000011.tar 256 0.0 GB 2816\n",
      "# writing data/vision/webdataset/train-000012.tar 256 0.0 GB 3072\n",
      "# writing data/vision/webdataset/train-000013.tar 256 0.0 GB 3328\n",
      "# writing data/vision/webdataset/train-000014.tar 256 0.0 GB 3584\n",
      "# writing data/vision/webdataset/train-000015.tar 256 0.0 GB 3840\n",
      "# writing data/vision/webdataset/train-000016.tar 256 0.0 GB 4096\n",
      "# writing data/vision/webdataset/train-000017.tar 256 0.0 GB 4352\n",
      "# writing data/vision/webdataset/train-000018.tar 256 0.0 GB 4608\n",
      "# writing data/vision/webdataset/train-000019.tar 256 0.0 GB 4864\n",
      "# writing data/vision/webdataset/train-000020.tar 256 0.0 GB 5120\n",
      "# writing data/vision/webdataset/train-000021.tar 256 0.0 GB 5376\n",
      "# writing data/vision/webdataset/train-000022.tar 256 0.0 GB 5632\n",
      "# writing data/vision/webdataset/train-000023.tar 256 0.0 GB 5888\n",
      "# writing data/vision/webdataset/train-000024.tar 256 0.0 GB 6144\n",
      "# writing data/vision/webdataset/train-000025.tar 256 0.0 GB 6400\n",
      "# writing data/vision/webdataset/train-000026.tar 256 0.0 GB 6656\n",
      "# writing data/vision/webdataset/train-000027.tar 256 0.0 GB 6912\n",
      "# writing data/vision/webdataset/train-000028.tar 256 0.0 GB 7168\n",
      "# writing data/vision/webdataset/train-000029.tar 256 0.0 GB 7424\n",
      "# writing data/vision/webdataset/train-000030.tar 256 0.0 GB 7680\n",
      "# writing data/vision/webdataset/train-000031.tar 256 0.0 GB 7936\n",
      "# writing data/vision/webdataset/train-000032.tar 256 0.0 GB 8192\n",
      "# writing data/vision/webdataset/train-000033.tar 256 0.0 GB 8448\n",
      "# writing data/vision/webdataset/train-000034.tar 256 0.0 GB 8704\n",
      "# writing data/vision/webdataset/train-000035.tar 256 0.0 GB 8960\n",
      "# writing data/vision/webdataset/train-000036.tar 256 0.0 GB 9216\n",
      "# writing data/vision/webdataset/train-000037.tar 256 0.0 GB 9472\n",
      "# writing data/vision/webdataset/train-000038.tar 256 0.0 GB 9728\n",
      "# writing data/vision/webdataset/train-000039.tar 256 0.0 GB 9984\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "import pickle\n",
    "import random\n",
    "import time\n",
    "\n",
    "import pandas as pd\n",
    "import webdataset as wds\n",
    "from tqdm.auto import tqdm\n",
    "\n",
    "\n",
    "def write_dogs_wd(root_dir='data/vision/dog-breed-identification', \n",
    "                  shard_dir='data/vision/webdataset',\n",
    "                  maxsize=1e9, maxcount=100000, shuffle=False):\n",
    "\n",
    "    assert maxsize > 10000000\n",
    "    assert maxcount < 1000000\n",
    "\n",
    "    if not os.path.exists(shard_dir):\n",
    "        os.makedirs(shard_dir)\n",
    "\n",
    "    root_dir = root_dir\n",
    "    image_dir = os.path.join(root_dir, 'train/')\n",
    "    img_list = pd.read_csv(os.path.join(root_dir, 'labels.csv'))\n",
    "    idx2label=list(img_list['breed'].unique())\n",
    "    label2idx = {b: i for i, b in enumerate(idx2label)}\n",
    "    nimages = img_list.shape[0]\n",
    "    print(f'nimages: {nimages}')\n",
    "    indexes = list(range(nimages))\n",
    "    if shuffle:\n",
    "        random.shuffle(indexes)\n",
    "\n",
    "    # This is the output pattern under which we write shards\n",
    "    pattern = os.path.join(shard_dir, f'train-%06d.tar')\n",
    "    with wds.ShardWriter(pattern, maxsize=int(maxsize), maxcount=int(maxcount), compress=False) as sink:\n",
    "\n",
    "        for i in tqdm(indexes):\n",
    "            key = f'{i:07d}'\n",
    "            img_row = img_list.iloc[i]\n",
    "            label = label2idx[img_row['breed']]\n",
    "            #label = label.to_bytes(1, 'big')\n",
    "            with open(os.path.join(image_dir, f\"{img_row['id']}.jpg\"), 'rb') as stream:\n",
    "                image = stream.read()\n",
    "\n",
    "            # Construct a sample\n",
    "            sample = {'__key__': key, 'jpg': image, 'cls': label}\n",
    "\n",
    "            # Write the sample to the sharded tar archives\n",
    "            sink.write(sample)\n",
    "\n",
    "    # Pickle class map\n",
    "    with open(os.path.join(shard_dir, 'idx2label'), 'wb') as fp:\n",
    "        pickle.dump(idx2label, fp)\n",
    "\n",
    "    return idx2label\n",
    "\n",
    "# Create dataset\n",
    "shard_dir = \"data/vision/webdataset\"\n",
    "idx2label = write_dogs_wd(maxcount=256, shuffle=False, shard_dir=shard_dir)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### DataLoading with the original format"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As a point of comparison, first we can check out what DataPipe construction with the non-convertered images and csv labels would look like. \n",
    "\n",
    "We'll run this next example as a script instead of directly in the notebook, since some multiprocessing functionality does not play well in an interactive environment."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Overwriting jpeg_images_csv_labels.py\n"
     ]
    }
   ],
   "source": [
    "%%writefile jpeg_images_csv_labels.py\n",
    "\n",
    "from pathlib import Path\n",
    "import torch\n",
    "import torchdata.datapipes as dp\n",
    "import torchvision.transforms as transforms\n",
    "#from torch.utils.data.datapipes.utils.decoder import imagehandler\n",
    "from PIL import Image, features\n",
    "import io\n",
    "import os\n",
    "from tqdm.auto import tqdm\n",
    "import pandas as pd\n",
    "from torchdata.dataloader2 import (DataLoader2,\n",
    "                                   PrototypeMultiProcessingReadingService)\n",
    "import PIL\n",
    "# Check that PIL-SIMD is installed with libjpeg-turbo\n",
    "assert \"post\" in PIL.__version__\n",
    "assert features.check_feature('libjpeg_turbo')\n",
    "\n",
    "\n",
    "device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
    "\n",
    "# Define simple transforms\n",
    "data_transform = transforms.Compose(\n",
    "    [\n",
    "        transforms.RandomResizedCrop(256),\n",
    "        transforms.RandomHorizontalFlip(),\n",
    "        transforms.RandomRotation((-45,45)),\n",
    "        transforms.ToTensor(),\n",
    "        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),\n",
    "    ]\n",
    ")\n",
    "\n",
    "# Define how to decode webdataset keys\n",
    "def decode(item, transform=data_transform):\n",
    "    key, value = item\n",
    "    if key.endswith(\"jpg\"):\n",
    "        image = Image.open(io.BytesIO(value.read()))\n",
    "        return key, transform(image)\n",
    "    else:\n",
    "        raise ValueError(f\"Unknown key: {key}\")\n",
    "\n",
    "ROOT = Path('data/vision/dog-breed-identification')\n",
    "\n",
    "# Here we load all the labels ahead of time\n",
    "labels = pd.read_csv(ROOT/'labels.csv')\n",
    "idx2label=list(labels['breed'].unique())\n",
    "label2idx = {b: i for i, b in enumerate(idx2label)}\n",
    "labels_dict = pd.Series(labels.breed.values,index=labels.id).to_dict()\n",
    "\n",
    "def collate_sample(data):\n",
    "    paths, images = zip(*data)\n",
    "    images = torch.stack(images).to(device)\n",
    "    labels = [labels_dict[os.path.basename(path).split('.')[0]] for path in paths]\n",
    "    classes = torch.Tensor([label2idx[label] for label in labels]).to(device)\n",
    "    return classes, images\n",
    "\n",
    "\n",
    "if __name__ == '__main__':\n",
    "\n",
    "    BATCH_SIZE = 256\n",
    "    NUM_WORKERS = 8\n",
    "\n",
    "    datapipe = dp.iter.FileLister([ROOT/'train',], masks=['*.jpg'])\n",
    "    datapipe = (datapipe\n",
    "                .shuffle()\n",
    "                .sharding_filter()\n",
    "                .open_files(mode=\"b\")\n",
    "                .map(decode)\n",
    "                .batch(BATCH_SIZE)\n",
    "                .collate(collate_sample)\n",
    "    )\n",
    "\n",
    "    reading_service = PrototypeMultiProcessingReadingService(\n",
    "        num_workers=NUM_WORKERS,\n",
    "        # To use CUDA with multiprocessing, you must use the 'spawn' start method\n",
    "        multiprocessing_context='spawn',\n",
    "    )\n",
    "\n",
    "    dl = DataLoader2(datapipe, reading_service=reading_service)\n",
    "\n",
    "    for batch in tqdm(dl, mininterval=0.5):\n",
    "        pass\n",
    "\n",
    "    print(batch[1][1].shape, batch[1][1].device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "40it [00:12,  3.08it/s]\n",
      "torch.Size([3, 256, 256]) cuda:0\n"
     ]
    }
   ],
   "source": [
    "!python jpeg_images_csv_labels.py"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Using the WebDataset DataPipe <a class=\"anchor\" id=\"26\"></a>\n",
    "\n",
    "Now here's an example of reading WebDataset Format:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Overwriting webdataset_with_multiprocessing.py\n"
     ]
    }
   ],
   "source": [
    "%%writefile webdataset_with_multiprocessing.py\n",
    "import io\n",
    "import os\n",
    "#import pickle\n",
    "from functools import partial\n",
    "\n",
    "import torch\n",
    "import torchvision.transforms as transforms\n",
    "from PIL import Image, features\n",
    "from tqdm.auto import tqdm\n",
    "#from webdataset.autodecode import imagehandler\n",
    "from torchdata.dataloader2 import (DataLoader2,\n",
    "                                   PrototypeMultiProcessingReadingService)\n",
    "from torchdata.datapipes.iter import FileLister, FileOpener\n",
    "import PIL\n",
    "# Check that PIL-SIMD is installed with libjpeg-turbo\n",
    "assert \"post\" in PIL.__version__\n",
    "assert features.check_feature('libjpeg_turbo')\n",
    "\n",
    "device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
    "\n",
    "# Define image decoders\n",
    "#torch_handler = imagehandler('torch')\n",
    "\n",
    "# Define simple transforms (since resizing is required to collate)\n",
    "data_transform = transforms.Compose(\n",
    "    [\n",
    "        transforms.RandomResizedCrop(256),\n",
    "        transforms.RandomHorizontalFlip(),\n",
    "        transforms.RandomRotation((-45,45)),\n",
    "        transforms.ToTensor(),\n",
    "        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),\n",
    "    ]\n",
    ")\n",
    "\n",
    "# Define how to decode webdataset keys\n",
    "def decode(item, transform=data_transform):\n",
    "    key, value = item\n",
    "    if key.endswith(\".cls\"):\n",
    "        class_id = int(value.read().decode(\"utf-8\"))\n",
    "        return key, class_id\n",
    "    elif key.endswith(\".jpg\"):\n",
    "        #image = image_decoder(\".jpg\", value.read())\n",
    "        image = Image.open(io.BytesIO(value.read()))\n",
    "        return key, transform(image)\n",
    "    else:\n",
    "        raise ValueError(f\"Unknown key: {key}\")\n",
    "\n",
    "# Define simple collate function\n",
    "def collate_fn(batch, device=device):\n",
    "    classes = torch.Tensor([sample['.cls'] for sample in batch]).to(device)\n",
    "    images = torch.stack([sample['.jpg'] for sample in batch]).to(device)\n",
    "    return classes, images\n",
    "\n",
    "\n",
    "if __name__ == '__main__':\n",
    "\n",
    "    BATCH_SIZE = 256\n",
    "    NUM_WORKERS = 8\n",
    "\n",
    "    # Create pipeline\n",
    "    shard_dir = \"data/vision/webdataset\"\n",
    "    datapipe = FileLister(shard_dir, \"train*.tar\")\n",
    "    dataset = (datapipe\n",
    "               .shuffle()\n",
    "               .sharding_filter()\n",
    "               .open_files(mode='b')\n",
    "               .load_from_tar()\n",
    "               .map(decode)\n",
    "               .webdataset()\n",
    "               .batch(BATCH_SIZE)\n",
    "               # stack lists of tensors and place on device\n",
    "               .collate(partial(collate_fn, device=device))\n",
    "    )\n",
    "\n",
    "    reading_service = PrototypeMultiProcessingReadingService(\n",
    "        num_workers=NUM_WORKERS,\n",
    "        # To use CUDA with multiprocessing, you must use the 'spawn' start method\n",
    "        multiprocessing_context='spawn',\n",
    "    )\n",
    "    dl = DataLoader2(dataset, reading_service=reading_service)\n",
    "\n",
    "    for batch in tqdm(dl, mininterval=0.5):\n",
    "        pass\n",
    "\n",
    "    print(batch[1][1].shape, batch[1][1].device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "40it [00:13,  2.91it/s]\n",
      "torch.Size([3, 256, 256]) cuda:0\n"
     ]
    }
   ],
   "source": [
    "# Run the example\n",
    "!python webdataset_with_multiprocessing.py"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Speeding up transforms with  Kornia <a class=\"anchor\" id=\"27\"></a>\n",
    "\n",
    "Here we'll move some of the augmentations to be performed batched on the GPU."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Writing faster_transforms.py\n"
     ]
    }
   ],
   "source": [
    "%%writefile faster_transforms.py\n",
    "import pickle\n",
    "import os\n",
    "import io\n",
    "\n",
    "import numpy as np\n",
    "import PIL\n",
    "from PIL import Image, features\n",
    "from tqdm.auto import tqdm\n",
    "from torchdata.datapipes.iter import FileLister, FileOpener\n",
    "from torchdata.dataloader2 import (DataLoader2,\n",
    "                                   PrototypeMultiProcessingReadingService)\n",
    "from webdataset.autodecode import imagehandler\n",
    "from functools import partial\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "\n",
    "from torchvision import transforms\n",
    "import kornia.augmentation as K\n",
    "\n",
    "# Check that PIL-SIMD is installed with libjpeg-turbo\n",
    "assert \"post\" in PIL.__version__\n",
    "assert features.check_feature('libjpeg_turbo')\n",
    "\n",
    "device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
    "\n",
    "# Define simple transforms (since resizing is required to collate)\n",
    "data_transform = transforms.Compose(\n",
    "    [\n",
    "        transforms.RandomResizedCrop(256),\n",
    "        transforms.ToTensor(),\n",
    "    ]\n",
    ")\n",
    "\n",
    "# Define how to decode webdataset keys\n",
    "def decode(item, transform=data_transform):\n",
    "    key, value = item\n",
    "    if key.endswith(\".cls\"):\n",
    "        class_id = int(value.read().decode(\"utf-8\"))\n",
    "        return key, class_id\n",
    "    elif key.endswith(\".jpg\"):\n",
    "        image = Image.open(io.BytesIO(value.read()))\n",
    "        return key, transform(image)\n",
    "    else:\n",
    "        raise ValueError(f\"Unknown key: {key}\")\n",
    "\n",
    "# Define simple collate function\n",
    "def collate_fn(batch, device=device):\n",
    "    classes = torch.Tensor([sample['.cls'] for sample in batch]).to(device)\n",
    "    images = torch.stack([sample['.jpg'] for sample in batch]).to(device)\n",
    "    return images, classes\n",
    "\n",
    "kornia_transform = nn.Sequential(\n",
    "    K.RandomHorizontalFlip(),\n",
    "    K.RandomRotation(degrees=45.0),\n",
    "    K.Normalize(mean=torch.Tensor([0.485, 0.456, 0.406])*255,\n",
    "                std=torch.Tensor([0.229, 0.224, 0.225])*255)\n",
    ")\n",
    "\n",
    "\n",
    "if __name__ == '__main__':\n",
    "\n",
    "    BATCH_SIZE = 256\n",
    "    NUM_WORKERS = 8\n",
    "\n",
    "    # Create pipeline\n",
    "    shard_dir = \"data/vision/webdataset\"\n",
    "    datapipe = FileLister(shard_dir, \"train*.tar\")\n",
    "    dataset = (datapipe\n",
    "              .shuffle()\n",
    "              .sharding_filter()\n",
    "              .open_files(mode='b')\n",
    "              .load_from_tar()\n",
    "              .map(decode)\n",
    "              .webdataset()\n",
    "              .batch(BATCH_SIZE)\n",
    "              # stack lists of tensors and place on device\n",
    "              .collate(partial(collate_fn, device=device))\n",
    "    )\n",
    "\n",
    "    reading_service = PrototypeMultiProcessingReadingService(\n",
    "        num_workers=NUM_WORKERS,\n",
    "        # To use CUDA with multiprocessing, you must use the 'spawn' start method\n",
    "        multiprocessing_context='spawn',\n",
    "    )\n",
    "    dl = DataLoader2(dataset, reading_service=reading_service)\n",
    "\n",
    "    for batch in tqdm(dl, mininterval=0.5):\n",
    "        images, classes = batch\n",
    "        images = kornia_transform(images)\n",
    "\n",
    "    print(images.shape, images.device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "40it [00:12,  3.26it/s]\n",
      "torch.Size([256, 3, 256, 256]) cuda:0\n"
     ]
    }
   ],
   "source": [
    "# Run the example\n",
    "!python faster_transforms.py"
   ]
  },
  {
   "attachments": {
    "image.png": {
     "image/png": "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"
    }
   },
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Some general observations regarding CV data loading\n",
    "- BGR to RGB conversion is slow, if you're using `cv2.imread`, just train with BGR\n",
    "- `cv2.imread` is slower than PIL-SIMD `Image.open`\n",
    "- It's a lot faster to perform normalization after converting to a `torch.Tensor`, even though it still works in the reverse order\n",
    "- Many image transformations can be done considerably on the GPU and batched (i.e. by using Kornia, DALI, or other libraries)\n",
    "- The bigger your images and the more augmentations you perform, the more likely I/O will be a bottleneck\n",
    "- The smaller your model the more likely I/O will be a bottleneck\n",
    "- Fastest speeds can often be achieved with NVIDIA DALI, but DALI is less versatile compared to TorchData, so I'd only recommend using it if you've identified I/O bottlenecks that necessitate its use\n",
    "    - DALI can perform batched resizing on the GPU\n",
    "\n",
    "To get an idea of the relative performance tradeoffs, here's an example of different dataloading methods, all with the same set of transforms (random resized crop, horizontal flip, vertical flip, random rotation, normalize):\n",
    "\n",
    "![image.png](attachment:image.png)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Object detection example <a class=\"anchor\" id=\"28\"></a>\n",
    "\n",
    "In this next example, we'll show how we can build a custom COCO datapipe using the original COCO format. However, in my opinion, converting COCO to webdataset is actually much cleaner. This example is adapted from [here](https://github.com/pytorch/vision/blob/7cf0f4cc1801ff1892007c7a11f7c35d8dfb7fd0/torchvision/prototype/datasets/_builtin/coco.py)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "A    data/vision/tiny_coco/instances_train2017.json\n",
      "A    data/vision/tiny_coco/train_2017_small\n",
      "A    data/vision/tiny_coco/train_2017_small/000000005802.jpg\n",
      "A    data/vision/tiny_coco/train_2017_small/000000060623.jpg\n",
      "A    data/vision/tiny_coco/train_2017_small/000000118113.jpg\n",
      "A    data/vision/tiny_coco/train_2017_small/000000184613.jpg\n",
      "A    data/vision/tiny_coco/train_2017_small/000000193271.jpg\n",
      "A    data/vision/tiny_coco/train_2017_small/000000222564.jpg\n",
      "A    data/vision/tiny_coco/train_2017_small/000000224736.jpg\n",
      "A    data/vision/tiny_coco/train_2017_small/000000309022.jpg\n",
      "A    data/vision/tiny_coco/train_2017_small/000000318219.jpg\n",
      "A    data/vision/tiny_coco/train_2017_small/000000374628.jpg\n",
      "A    data/vision/tiny_coco/train_2017_small/000000391895.jpg\n",
      "A    data/vision/tiny_coco/train_2017_small/000000403013.jpg\n",
      "A    data/vision/tiny_coco/train_2017_small/000000483108.jpg\n",
      "A    data/vision/tiny_coco/train_2017_small/000000522418.jpg\n",
      "A    data/vision/tiny_coco/train_2017_small/000000554625.jpg\n",
      "A    data/vision/tiny_coco/train_2017_small/000000574769.jpg\n",
      "Checked out revision 5.\n",
      "instances_train2017.json  train_2017_small\n"
     ]
    }
   ],
   "source": [
    "# Download the data\n",
    "!mkdir -p data/vision/tiny_coco\n",
    "!svn checkout https://github.com/austinmw/tiny-coco.git/trunk/tiny_coco data/vision/tiny_coco\n",
    "!ls data/vision/tiny_coco"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'path': '/home/ec2-user/SageMaker/torchdata-examples/data/vision/tiny_coco/train_2017_small/000000554625.jpg',\n",
       " 'image': EncodedImage([255, 216, 255,  ..., 194, 255, 217], dtype=torch.uint8),\n",
       " 'segmentations': Mask([[[False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        ...,\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False]],\n",
       " \n",
       "       [[False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        ...,\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False]],\n",
       " \n",
       "       [[False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        ...,\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False]],\n",
       " \n",
       "       ...,\n",
       " \n",
       "       [[False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        ...,\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False]],\n",
       " \n",
       "       [[False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        ...,\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False]],\n",
       " \n",
       "       [[False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        ...,\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False],\n",
       "        [False, False, False,  ..., False, False, False]]]),\n",
       " 'areas': Datapoint([ 7821.5317,  3031.9434, 29506.2734,  6283.4087,  3514.0569,\n",
       "             4582.3252,  1333.0531,   867.6927,  7658.1880,  3955.6077,\n",
       "             2218.4204,  3604.6162,   230.4866,   617.1909, 74480.1328,\n",
       "             3659.0964,  1192.4875,   147.5133,   929.2119]),\n",
       " 'crowds': Datapoint([False, False, False, False, False, False, False, False, False, False,\n",
       "            False, False, False, False, False, False, False, False, False]),\n",
       " 'bounding_boxes': BoundingBox([[3.8074e+02, 1.1285e+02, 4.0620e+01, 2.4882e+02],\n",
       "              [3.3913e+02, 3.2990e+01, 3.2990e+01, 1.7550e+02],\n",
       "              [2.8700e+00, 7.3180e+01, 1.4350e+02, 5.3238e+02],\n",
       "              [1.0520e+02, 3.6830e+01, 1.1206e+02, 2.2787e+02],\n",
       "              [8.4030e+01, 1.5700e+00, 1.1094e+02, 6.4990e+01],\n",
       "              [3.5955e+02, 6.8790e+01, 4.6020e+01, 2.0279e+02],\n",
       "              [3.0272e+02, 5.8032e+02, 6.9850e+01, 4.7150e+01],\n",
       "              [3.3780e+02, 4.2855e+02, 5.6300e+01, 2.3310e+01],\n",
       "              [2.4333e+02, 5.0530e+02, 1.2690e+02, 7.6410e+01],\n",
       "              [2.2733e+02, 3.9725e+02, 1.1144e+02, 5.3210e+01],\n",
       "              [2.1426e+02, 3.2190e+02, 7.2500e+01, 3.7750e+01],\n",
       "              [3.2248e+02, 0.0000e+00, 4.7250e+01, 1.5832e+02],\n",
       "              [2.2102e+02, 3.8268e+02, 6.9390e+01, 1.4450e+01],\n",
       "              [3.0878e+02, 4.7052e+02, 8.4580e+01, 2.9170e+01],\n",
       "              [2.7330e+01, 1.6683e+02, 3.1928e+02, 4.7317e+02],\n",
       "              [5.9990e+01, 5.5160e+01, 1.0153e+02, 1.1956e+02],\n",
       "              [3.0661e+02, 3.3000e-01, 2.8950e+01, 7.1640e+01],\n",
       "              [2.8453e+02, 3.1254e+02, 2.0870e+01, 8.3500e+00],\n",
       "              [2.9489e+02, 2.8149e+02, 4.3990e+01, 3.1840e+01]], format=BoundingBoxFormat.XYWH, spatial_size=(640, 426)),\n",
       " 'labels': Label([72, 72,  1,  1,  1, 72, 74, 74, 76, 76, 76, 72, 74, 74,  1,  1, 72, 74,\n",
       "        76]),\n",
       " 'super_categories': ['electronic',\n",
       "  'electronic',\n",
       "  'person',\n",
       "  'person',\n",
       "  'person',\n",
       "  'electronic',\n",
       "  'electronic',\n",
       "  'electronic',\n",
       "  'electronic',\n",
       "  'electronic',\n",
       "  'electronic',\n",
       "  'electronic',\n",
       "  'electronic',\n",
       "  'electronic',\n",
       "  'person',\n",
       "  'person',\n",
       "  'electronic',\n",
       "  'electronic',\n",
       "  'electronic'],\n",
       " 'ann_ids': [30093,\n",
       "  35249,\n",
       "  200698,\n",
       "  215463,\n",
       "  231321,\n",
       "  1097584,\n",
       "  1107477,\n",
       "  1107582,\n",
       "  1115813,\n",
       "  1117102,\n",
       "  1117343,\n",
       "  1624922,\n",
       "  1629004,\n",
       "  1629550,\n",
       "  1708274,\n",
       "  1726997,\n",
       "  1969231,\n",
       "  1971582,\n",
       "  1975400]}"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import os\n",
    "import csv\n",
    "import requests\n",
    "from typing import Any, List, Union, Sequence, Dict\n",
    "\n",
    "from torchvision.prototype import datasets\n",
    "from torchvision.prototype.datasets import register_info, register_dataset, Coco\n",
    "from torchvision.prototype.datasets.utils import ManualDownloadResource\n",
    "\n",
    "NAME = \"tiny_coco\"\n",
    "\n",
    "@register_info(NAME)\n",
    "def _info() -> Dict[str, Any]:\n",
    "    return datasets.info('coco')\n",
    "\n",
    "@register_dataset(NAME)\n",
    "class TinyCoco(Coco):\n",
    "    def _resources(self) -> List[ManualDownloadResource]:\n",
    "        images_resource = ManualDownloadResource(\n",
    "            file_name=\"train_2017_small\", \n",
    "            instructions=\"Download from https://github.com/austinmw/tiny-coco\"\n",
    "        )\n",
    "        meta_resource = ManualDownloadResource(\n",
    "            file_name=\"instances_train2017.json\", \n",
    "            instructions=\"Download from https://github.com/austinmw/tiny-coco\"\n",
    "        )\n",
    "        resource_dp = [images_resource, meta_resource]\n",
    "        return resource_dp\n",
    "\n",
    "\n",
    "assert NAME in datasets.list_datasets()\n",
    "\n",
    "dp = datasets.load(\"tiny_coco\", root=\"data/vision/tiny_coco\")\n",
    "next(iter(dp))"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Language Data Examples <a class=\"anchor\" id=\"29\"></a>\n",
    "\n",
    "### TorchText with DataLoader2 [[11]](https://github.com/pytorch/data/blob/main/examples/dataloader2/train_loop_torchtext.py) <a class=\"anchor\" id=\"30\"></a>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "4115466b3ab64afcaa8175438c2ac149",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import torch\n",
    "from torch.hub import load_state_dict_from_url\n",
    "from torchdata.dataloader2 import DataLoader2\n",
    "import torchtext\n",
    "import torchtext.functional as F\n",
    "import torchtext.transforms as T\n",
    "from torchtext.datasets import SST2\n",
    "from tqdm.auto import tqdm\n",
    "\n",
    "PADDING_IDX = 1\n",
    "BOS_IDX = 0\n",
    "EOS_IDX = 2\n",
    "MAX_SEQ_LEN = 256\n",
    "NUM_EPOCHS = 1\n",
    "BATCH_SIZE = 8\n",
    "XLMR_VOCAB_PATH = r\"https://download.pytorch.org/models/text/xlmr.vocab.pt\"\n",
    "XLMR_SPM_MODEL_PATH = r\"https://download.pytorch.org/models/text/xlmr.sentencepiece.bpe.model\"\n",
    "\n",
    "text_transform = T.Sequential(\n",
    "    T.SentencePieceTokenizer(XLMR_SPM_MODEL_PATH),\n",
    "    T.VocabTransform(load_state_dict_from_url(XLMR_VOCAB_PATH)),\n",
    "    T.Truncate(MAX_SEQ_LEN - 2),\n",
    "    T.AddToken(token=BOS_IDX, begin=True),\n",
    "    T.AddToken(token=EOS_IDX, begin=False),\n",
    ")\n",
    "\n",
    "def apply_transform(x):\n",
    "    return text_transform(x[0]), x[1]\n",
    "\n",
    "train_datapipe = SST2(split=\"train\")\n",
    "train_datapipe = train_datapipe.map(apply_transform)\n",
    "train_datapipe = train_datapipe.batch(BATCH_SIZE)\n",
    "train_datapipe = train_datapipe.rows2columnar([\"token_ids\", \"target\"])\n",
    "train_dataloader = DataLoader2(datapipe=train_datapipe)\n",
    "\n",
    "DEVICE = torch.device(\"cuda\") if torch.cuda.is_available() else torch.device(\"cpu\")\n",
    "\n",
    "for epoch in range(NUM_EPOCHS):\n",
    "    for step, batch in tqdm(enumerate(train_dataloader), maxinterval=0.5):\n",
    "        input = F.to_tensor(batch[\"token_ids\"], padding_value=PADDING_IDX).to(DEVICE)\n",
    "        target = torch.tensor(batch[\"target\"]).to(DEVICE)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### HuggingFace Datasets <a class=\"anchor\" id=\"31\"></a>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using custom data configuration lhoestq--demo1-cb69bfa340c0f55f\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'id': '7bd227d9-afc9-11e6-aba1-c4b301cdf627',\n",
       " 'package_name': 'com.mantz_it.rfanalyzer',\n",
       " 'review': \"Great app! The new version now works on my Bravia Android TV which is great as it's right by my rooftop aerial cable. The scan feature would be useful...any ETA on when this will be available? Also the option to import a list of bookmarks e.g. from a simple properties file would be useful.\",\n",
       " 'date': 'October 12 2016',\n",
       " 'star': 4,\n",
       " 'version_id': 1487}"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from torchdata.datapipes.iter import HuggingFaceHubReader\n",
    "\n",
    "datapipe = HuggingFaceHubReader(dataset=\"lhoestq/demo1\", split=\"train\", revision=\"main\", streaming=True) #, use_auth_token=True)\n",
    "\n",
    "next(iter(datapipe))"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----\n",
    "\n",
    "## Custom DataPipes [[12]](https://pytorch.org/data/beta/tutorial.html#implementing-a-custom-datapipe) <a class=\"anchor\" id=\"32\"></a>\n",
    "\n",
    "If none of the built-in DataPipes supports your need, you can create your own custom DataPipe.\n",
    "\n",
    "As a guiding example, let us implement an IterDataPipe that applies a callable to the input iterator."
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Naming\n",
    "\n",
    "The naming convention for `DataPipe` is “Operation”-er, followed by `IterDataPipe` or `MapDataPipe`, as each `DataPipe` is essentially a container to apply an operation to data yielded from a source `DataPipe`. For succinctness, we alias to just “Operation-er” in init files. For our `IterDataPipe` example, we’ll name the module `MapperIterDataPipe` and alias it as `iter.Mapper` under `torchdata.datapipes`.\n",
    "\n",
    "For the functional method name, the naming convention is `datapipe.<operation>`. For instance, the functional method name of `Mapper` is `map`, such that it can be invoked by `datapipe.map(...)`."
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Constructor\n",
    "\n",
    "DataSets are now generally constructed as stacks of DataPipes, so each `DataPipe` typically takes a source `DataPipe` as its first argument. Here is a simplified version of `Mapper` as an example:\n",
    "\n",
    "```python\n",
    "    from torchdata.datapipes.iter import IterDataPipe\n",
    "\n",
    "    class MapperIterDataPipe(IterDataPipe):\n",
    "        def __init__(self, source_dp: IterDataPipe, fn) -> None:\n",
    "            super().__init__()\n",
    "            self.source_dp = source_dp\n",
    "            self.fn = fn\n",
    "```"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note:\n",
    "\n",
    "- Avoid loading data from the source DataPipe in `__init__` function, in order to support lazy data loading and save memory.\n",
    "- If `IterDataPipe` instance holds data in memory, please be aware of the in-place modification of data. When second iterator is created from the instance, the data may have already changed. Please take `IterableWrapper` class as reference to `deepcopy` data for each iterator.\n",
    "- Avoid variables names that are taken by the functional names of existing DataPipes. For instance, `.filter` is the functional name that can be used to invoke `FilterIterDataPipe`. Having a variable named `filter` inside another `IterDataPipe` can lead to confusion."
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Iterator\n",
    "\n",
    "For `IterDataPipes`, an `__iter__` function is needed to consume data from the source `IterDataPipe` then apply the operation over the data before `yield`.\n",
    "\n",
    "```python\n",
    "    class MapperIterDataPipe(IterDataPipe):\n",
    "        # ... See __init__() defined above\n",
    "\n",
    "        def __iter__(self):\n",
    "            for d in self.dp:\n",
    "                yield self.fn(d)\n",
    "```"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Length\n",
    "\n",
    "In many cases, as in our `MapperIterDataPipe` example, the `__len__` method of a DataPipe returns the length of the source DataPipe.\n",
    "\n",
    "```python\n",
    "    class MapperIterDataPipe(IterDataPipe):\n",
    "        # ... See __iter__() defined above\n",
    "\n",
    "        def __len__(self):\n",
    "            return len(self.dp)\n",
    "```\n",
    "\n",
    "However, note that `__len__` is optional for IterDataPipe and often inadvisable. For `CSVParserIterDataPipe` in the [Using DataPipes](#10) section, `__len__` is not implemented because the number of rows in each file is unknown before loading it. In some special cases, `__len__` can be made to either return an integer or raise an Error depending on the input. In those cases, the Error must be a `TypeError` to support Python’s build-in functions like `list(dp)`."
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Registering DataPipes with the functional API\n",
    "\n",
    "Each DataPipe can be registered to support functional invocation using the decorator `functional_datapipe`.\n",
    "\n",
    "```python\n",
    "   from torchdata.datapipes import functional_datapipe\n",
    "\n",
    "   @functional_datapipe(\"map\")\n",
    "   class MapperIterDataPipe(IterDataPipe):\n",
    "   # ...\n",
    "```\n",
    "\n",
    "The stack of DataPipes can then be constructed using their functional forms (recommended) or class constructors:\n",
    "\n",
    "```python\n",
    "   import torchdata.datapipes as dp\n",
    "\n",
    "   # Using functional form (recommended)\n",
    "   datapipes1 = dp.iter.FileOpener(['a.file', 'b.file']).map(fn=decoder).shuffle().batch(2)\n",
    "   # Using class constructors\n",
    "   datapipes2 = dp.iter.FileOpener(['a.file', 'b.file'])\n",
    "   datapipes2 = dp.iter.Mapper(datapipes2, fn=decoder)\n",
    "   datapipes2 = dp.iter.Shuffler(datapipes2)\n",
    "   datapipes2 = dp.iter.Batcher(datapipes2, 2)\n",
    "```\n",
    "\n",
    "In the above example, `datapipes1` and `datapipes2` represent the exact same stack of IterDataPipes. We recommend using the functional form of DataPipes."
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----\n",
    "\n",
    "## Distributed training <a class=\"anchor\" id=\"33\"></a>\n",
    "\n",
    "A feature [currently in development](https://github.com/pytorch/data/issues/911) called `SequentialReadingService` will soon combine the features of `DistributedReadingService` and `MultiprocessingReadingService`. '\n",
    "\n",
    "As a workaround for now, all distributed features have been added to to [`PrototypeMultiProcessingReadingService`](https://github.com/pytorch/data/blob/main/torchdata/dataloader2/reading_service.py#L150) (except attaching `fullsync` operation to the pipeline). Therefore for now dist+mp can be achieved by:\n",
    "\n",
    "```python\n",
    "    rs = PrototypeMultiProcessingReadingService(...)\n",
    "    dl = DataLoader2(dp, reading_service=rs)\n",
    "    dp = IterableWrapper(dl).fullsync()  # Attach FullSync to the original pipeline\n",
    "```\n",
    "\n",
    "To achieve [distributed training without MP currently](https://github.com/pytorch/data/issues/689#issuecomment-1378967719), you can do:\n",
    "\n",
    "```python\n",
    "    datapipe = source_dp.shuffle().sharding_filter()  # Make sure distributed sharding is mutually exclusive and collectively exhaustive\n",
    "    datapipe = datapipe.map(transform_fn)\n",
    "    ...\n",
    "    def main(rank, world_size)\n",
    "        dist.init_process_group(rank=rank, world_size=world_size)\n",
    "        rs = DistributedReadingService()\n",
    "        dl = DataLoader2(datapipe, reading_service=rs)\n",
    "        for epoch in range(10):\n",
    "            torch.manual_seed(epoch)  # Set determinism for all ranks\n",
    "            for d in dl:\n",
    "                ...\n",
    "```\n",
    "\n",
    "### With SageMaker <a class=\"anchor\" id=\"34\"></a>\n",
    "\n",
    "A SageMaker example is below. The script `sagemaker/sagemaker_distributed_reading_service.py` was adapted from a TorchData example [here](https://github.com/pytorch/data/blob/main/examples/dataloader2/train_loop_distributed_reading_service.py)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "!mkdir -p sagemaker"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Writing sagemaker/requirements.txt\n"
     ]
    }
   ],
   "source": [
    "%%writefile sagemaker/requirements.txt\n",
    "torchdata==0.5.1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Overwriting sagemaker/sagemaker_distributed_reading_service.py\n"
     ]
    }
   ],
   "source": [
    "%%writefile sagemaker/sagemaker_distributed_reading_service.py\n",
    "import os\n",
    "\n",
    "import torch\n",
    "import smdistributed.dataparallel.torch.torch_smddp\n",
    "import torch.distributed as dist\n",
    "from torch.nn.parallel import DistributedDataParallel as DDP\n",
    "from torch import nn\n",
    "\n",
    "from torchdata.dataloader2 import DataLoader2, DistributedReadingService\n",
    "from torchdata.datapipes.iter import IterableWrapper\n",
    "\n",
    "\n",
    "class ToyModel(nn.Module):\n",
    "    def __init__(self) -> None:\n",
    "        \"\"\"\n",
    "        In the model constructor, we instantiate four parameters and use them\n",
    "        as member parameters.\n",
    "        \"\"\"\n",
    "        super().__init__()\n",
    "        self.a = nn.Parameter(torch.randn(()))\n",
    "        self.b = nn.Parameter(torch.randn(()))\n",
    "        self.c = nn.Parameter(torch.randn(()))\n",
    "        self.d = nn.Parameter(torch.randn(()))\n",
    "\n",
    "    def forward(self, x: torch.Tensor) -> torch.Tensor:\n",
    "        \"\"\"\n",
    "        Simple model forward function\n",
    "        \"\"\"\n",
    "        return self.a + self.b * x + self.c * x ** 2 + self.d * x ** 3\n",
    "\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "\n",
    "    dist.init_process_group(\"smddp\")\n",
    "    world_size = dist.get_world_size()\n",
    "    rank = dist.get_rank()\n",
    "    local_rank = int(os.getenv(\"LOCAL_RANK\", -1))\n",
    "\n",
    "    #os.environ[\"RANK\"] = str(0)\n",
    "    #os.environ[\"WORLD_SIZE\"] = str(2)\n",
    "    #os.environ[\"MASTER_ADDR\"] = \"localhost\"\n",
    "    #os.environ[\"MASTER_PORT\"] = \"0\"\n",
    "\n",
    "    batch_size = 32\n",
    "    batch_size //= world_size // 8\n",
    "    batch_size = max(batch_size, 1)\n",
    "\n",
    "    device = torch.device(\"cuda\")\n",
    "\n",
    "    model = ToyModel()\n",
    "    model = model.to(device)\n",
    "    model = DDP(model)\n",
    "\n",
    "    # SageMaker data parallel: Pin each GPU to a single library process.\n",
    "    torch.cuda.set_device(local_rank)\n",
    "    model.cuda(local_rank)\n",
    "\n",
    "    # Use a prime number to make sure uneven data sharding and let\n",
    "    # DistributedReadingService prevent hanging with the unbalanced data shard\n",
    "    data_length = 19997\n",
    "\n",
    "    train_features = IterableWrapper([torch.rand(3) for _ in range(data_length)])\n",
    "    train_labels = IterableWrapper([torch.rand(3) for _ in range(data_length)])\n",
    "\n",
    "    # sharding_filter will automatically shard the data based on the\n",
    "    # distributed ranks\n",
    "    train_data_pipe = train_features.zip(train_labels).shuffle().sharding_filter()\n",
    "\n",
    "    # Torch Distributed is required to use DistributedReadingService\n",
    "    reading_service = DistributedReadingService()\n",
    "    # To track SequentialReadingService: https://github.com/pytorch/data/issues/911\n",
    "    #reading_service = PrototypeMultiProcessingReadingService(num_workers=8)\n",
    "\n",
    "    # Create DataLoader2 with DistributedReadingService\n",
    "    data_loader2 = DataLoader2(\n",
    "        datapipe=train_data_pipe,\n",
    "        reading_service=reading_service,\n",
    "    )\n",
    "\n",
    "    criterion = torch.nn.MSELoss(reduction=\"sum\")\n",
    "    optimizer = torch.optim.SGD(model.parameters(), lr=1e-7)\n",
    "\n",
    "    for epoch in range(2):\n",
    "\n",
    "        # Set manual seed per epoch to control the randomness for shuffle.\n",
    "        torch.manual_seed(epoch)\n",
    "\n",
    "        running_loss = 0.0\n",
    "        for step, data in enumerate(data_loader2):\n",
    "            train_feature, train_label = data\n",
    "            train_feature = train_feature.to(device)\n",
    "            train_label = train_label.to(device)\n",
    "            optimizer.zero_grad()\n",
    "\n",
    "            predicted_outputs = model(train_feature)\n",
    "            loss = criterion(predicted_outputs, train_label)\n",
    "            loss.backward()\n",
    "            optimizer.step()\n",
    "\n",
    "            running_loss += loss.item()\n",
    "            if rank == 0 and step % 20 == 0:\n",
    "                print(\"[epoch: %d, %5d] loss: %.3f\" % (epoch + 1, step + 1, running_loss / 2000))\n",
    "            running_loss = 0.0\n",
    "\n",
    "        # SageMaker data parallel: Save model on master node.\n",
    "        if rank == 0:\n",
    "            torch.save(model.state_dict(), \"/opt/ml/model/model.pth\")\n",
    "\n",
    "    if rank == 0:\n",
    "        print(\"Finished Training\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:sagemaker:Creating training-job with name: pytorch-training-2023-02-02-20-42-59-071\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2023-02-02 20:42:59 Starting - Starting the training job......\n",
      "2023-02-02 20:43:56 Starting - Preparing the instances for training......\n",
      "2023-02-02 20:44:55 Downloading - Downloading input data...\n",
      "2023-02-02 20:45:15 Training - Downloading the training image.....................\n",
      "2023-02-02 20:48:51 Training - Training image download completed. Training in progress....bash: cannot set terminal process group (-1): Inappropriate ioctl for device\n",
      "bash: no job control in this shell\n",
      "2023-02-02 20:49:19,376 sagemaker-training-toolkit INFO     Imported framework sagemaker_pytorch_container.training\n",
      "2023-02-02 20:49:19,441 sagemaker-training-toolkit INFO     No Neurons detected (normal if no neurons installed)\n",
      "2023-02-02 20:49:19,454 sagemaker_pytorch_container.training INFO     Block until all host DNS lookups succeed.\n",
      "2023-02-02 20:49:19,457 sagemaker_pytorch_container.training INFO     Invoking SMDataParallel\n",
      "2023-02-02 20:49:19,457 sagemaker_pytorch_container.training INFO     Invoking user training script.\n",
      "2023-02-02 20:49:19,705 sagemaker-training-toolkit INFO     Installing dependencies from requirements.txt:\n",
      "/opt/conda/bin/python3.9 -m pip install -r requirements.txt\n",
      "Requirement already satisfied: torchdata==0.5.1 in /opt/conda/lib/python3.9/site-packages (from -r requirements.txt (line 1)) (0.5.1)\n",
      "Requirement already satisfied: urllib3>=1.25 in /opt/conda/lib/python3.9/site-packages (from torchdata==0.5.1->-r requirements.txt (line 1)) (1.26.13)\n",
      "Requirement already satisfied: portalocker>=2.0.0 in /opt/conda/lib/python3.9/site-packages (from torchdata==0.5.1->-r requirements.txt (line 1)) (2.7.0)\n",
      "Requirement already satisfied: torch==1.13.1 in /opt/conda/lib/python3.9/site-packages (from torchdata==0.5.1->-r requirements.txt (line 1)) (1.13.1+cu117)\n",
      "Requirement already satisfied: requests in /opt/conda/lib/python3.9/site-packages (from torchdata==0.5.1->-r requirements.txt (line 1)) (2.28.2)\n",
      "Requirement already satisfied: typing-extensions in /opt/conda/lib/python3.9/site-packages (from torch==1.13.1->torchdata==0.5.1->-r requirements.txt (line 1)) (4.4.0)\n",
      "Requirement already satisfied: charset-normalizer<4,>=2 in /opt/conda/lib/python3.9/site-packages (from requests->torchdata==0.5.1->-r requirements.txt (line 1)) (2.1.1)\n",
      "Requirement already satisfied: certifi>=2017.4.17 in /opt/conda/lib/python3.9/site-packages (from requests->torchdata==0.5.1->-r requirements.txt (line 1)) (2022.12.7)\n",
      "Requirement already satisfied: idna<4,>=2.5 in /opt/conda/lib/python3.9/site-packages (from requests->torchdata==0.5.1->-r requirements.txt (line 1)) (3.4)\n",
      "WARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\n",
      "2023-02-02 20:49:21,813 sagemaker-training-toolkit INFO     Waiting for the process to finish and give a return code.\n",
      "2023-02-02 20:49:21,813 sagemaker-training-toolkit INFO     Done waiting for a return code. Received 0 from exiting process.\n",
      "2023-02-02 20:49:21,881 sagemaker-training-toolkit INFO     No Neurons detected (normal if no neurons installed)\n",
      "2023-02-02 20:49:21,960 sagemaker-training-toolkit INFO     No Neurons detected (normal if no neurons installed)\n",
      "2023-02-02 20:49:21,973 sagemaker-training-toolkit INFO     Starting MPI run as worker node.\n",
      "2023-02-02 20:49:21,973 sagemaker-training-toolkit INFO     Creating SSH daemon.\n",
      "2023-02-02 20:49:21,975 sagemaker-training-toolkit INFO     Waiting for MPI workers to establish their SSH connections\n",
      "2023-02-02 20:49:21,977 sagemaker-training-toolkit INFO     Cannot connect to host algo-2 at port 22. Retrying...\n",
      "2023-02-02 20:49:21,977 sagemaker-training-toolkit INFO     Connection closed\n",
      "2023-02-02 20:49:22,978 sagemaker-training-toolkit INFO     Cannot connect to host algo-2 at port 22. Retrying...\n",
      "2023-02-02 20:49:22,978 sagemaker-training-toolkit INFO     Connection closed\n",
      "bash: cannot set terminal process group (-1): Inappropriate ioctl for device\n",
      "bash: no job control in this shell\n",
      "2023-02-02 20:49:20,340 sagemaker-training-toolkit INFO     Imported framework sagemaker_pytorch_container.training\n",
      "2023-02-02 20:49:20,405 sagemaker-training-toolkit INFO     No Neurons detected (normal if no neurons installed)\n",
      "2023-02-02 20:49:20,418 sagemaker_pytorch_container.training INFO     Block until all host DNS lookups succeed.\n",
      "2023-02-02 20:49:20,420 sagemaker_pytorch_container.training INFO     Invoking SMDataParallel\n",
      "2023-02-02 20:49:20,421 sagemaker_pytorch_container.training INFO     Invoking user training script.\n",
      "2023-02-02 20:49:20,662 sagemaker-training-toolkit INFO     Installing dependencies from requirements.txt:\n",
      "/opt/conda/bin/python3.9 -m pip install -r requirements.txt\n",
      "Requirement already satisfied: torchdata==0.5.1 in /opt/conda/lib/python3.9/site-packages (from -r requirements.txt (line 1)) (0.5.1)\n",
      "Requirement already satisfied: requests in /opt/conda/lib/python3.9/site-packages (from torchdata==0.5.1->-r requirements.txt (line 1)) (2.28.2)\n",
      "Requirement already satisfied: urllib3>=1.25 in /opt/conda/lib/python3.9/site-packages (from torchdata==0.5.1->-r requirements.txt (line 1)) (1.26.13)\n",
      "Requirement already satisfied: torch==1.13.1 in /opt/conda/lib/python3.9/site-packages (from torchdata==0.5.1->-r requirements.txt (line 1)) (1.13.1+cu117)\n",
      "Requirement already satisfied: portalocker>=2.0.0 in /opt/conda/lib/python3.9/site-packages (from torchdata==0.5.1->-r requirements.txt (line 1)) (2.7.0)\n",
      "Requirement already satisfied: typing-extensions in /opt/conda/lib/python3.9/site-packages (from torch==1.13.1->torchdata==0.5.1->-r requirements.txt (line 1)) (4.4.0)\n",
      "Requirement already satisfied: certifi>=2017.4.17 in /opt/conda/lib/python3.9/site-packages (from requests->torchdata==0.5.1->-r requirements.txt (line 1)) (2022.12.7)\n",
      "Requirement already satisfied: charset-normalizer<4,>=2 in /opt/conda/lib/python3.9/site-packages (from requests->torchdata==0.5.1->-r requirements.txt (line 1)) (2.1.1)\n",
      "Requirement already satisfied: idna<4,>=2.5 in /opt/conda/lib/python3.9/site-packages (from requests->torchdata==0.5.1->-r requirements.txt (line 1)) (3.4)\n",
      "WARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\n",
      "2023-02-02 20:49:22,738 sagemaker-training-toolkit INFO     Waiting for the process to finish and give a return code.\n",
      "2023-02-02 20:49:22,739 sagemaker-training-toolkit INFO     Done waiting for a return code. Received 0 from exiting process.\n",
      "2023-02-02 20:49:22,807 sagemaker-training-toolkit INFO     No Neurons detected (normal if no neurons installed)\n",
      "2023-02-02 20:49:22,886 sagemaker-training-toolkit INFO     No Neurons detected (normal if no neurons installed)\n",
      "2023-02-02 20:49:22,899 sagemaker-training-toolkit INFO     Starting MPI run as worker node.\n",
      "2023-02-02 20:49:22,899 sagemaker-training-toolkit INFO     Waiting for MPI Master to create SSH daemon.\n",
      "2023-02-02 20:49:22,912 paramiko.transport INFO     Connected (version 2.0, client OpenSSH_8.2p1)\n",
      "2023-02-02 20:49:23,086 paramiko.transport INFO     Authentication (publickey) successful!\n",
      "2023-02-02 20:49:23,086 sagemaker-training-toolkit INFO     Can connect to host algo-1\n",
      "2023-02-02 20:49:23,086 sagemaker-training-toolkit INFO     MPI Master online, creating SSH daemon.\n",
      "2023-02-02 20:49:23,086 sagemaker-training-toolkit INFO     Writing environment variables to /etc/environment for the MPI process.\n",
      "2023-02-02 20:49:23,090 sagemaker-training-toolkit INFO     Waiting for MPI process to finish.\n",
      "2023-02-02 20:49:23,992 paramiko.transport INFO     Connected (version 2.0, client OpenSSH_8.2p1)\n",
      "2023-02-02 20:49:24,170 paramiko.transport INFO     Authentication (publickey) successful!\n",
      "2023-02-02 20:49:24,171 sagemaker-training-toolkit INFO     Can connect to host algo-2 at port 22\n",
      "2023-02-02 20:49:24,171 sagemaker-training-toolkit INFO     Connection closed\n",
      "2023-02-02 20:49:24,171 sagemaker-training-toolkit INFO     Worker algo-2 available for communication\n",
      "2023-02-02 20:49:24,171 sagemaker-training-toolkit INFO     Network interface name: eth0\n",
      "2023-02-02 20:49:24,171 sagemaker-training-toolkit INFO     Host: ['algo-1', 'algo-2']\n",
      "2023-02-02 20:49:24,172 sagemaker-training-toolkit INFO     sagemaker_communication_backend: None\n",
      "2023-02-02 20:49:24,172 sagemaker-training-toolkit WARNING  Missing library /opt/conda/lib/libsmddp.so for SMDDP collective\n",
      "2023-02-02 20:49:24,172 sagemaker-training-toolkit WARNING  The system is not configured to run SMDDP collectives optimizedfor AWS infrastructure.Please use the latest SageMaker Deep Learning Container (DLC) to enable SMDDP Collectives support.\n",
      "Continuing model training with default NCCL communication backend.\n",
      "2023-02-02 20:49:24,172 sagemaker-training-toolkit INFO     instance type: ml.p3.16xlarge\n",
      "2023-02-02 20:49:24,172 sagemaker-training-toolkit INFO     Env Hosts: ['algo-1', 'algo-2'] Hosts: ['algo-1:8', 'algo-2:8'] process_per_hosts: 8 num_processes: 16\n",
      "2023-02-02 20:49:24,247 sagemaker-training-toolkit INFO     No Neurons detected (normal if no neurons installed)\n",
      "2023-02-02 20:49:24,261 sagemaker-training-toolkit INFO     Invoking user script\n",
      "Training Env:\n",
      "{\n",
      "    \"additional_framework_parameters\": {\n",
      "        \"sagemaker_distributed_dataparallel_custom_mpi_options\": \"\",\n",
      "        \"sagemaker_distributed_dataparallel_enabled\": true,\n",
      "        \"sagemaker_instance_type\": \"ml.p3.16xlarge\"\n",
      "    },\n",
      "    \"channel_input_dirs\": {},\n",
      "    \"current_host\": \"algo-1\",\n",
      "    \"current_instance_group\": \"homogeneousCluster\",\n",
      "    \"current_instance_group_hosts\": [\n",
      "        \"algo-1\",\n",
      "        \"algo-2\"\n",
      "    ],\n",
      "    \"current_instance_type\": \"ml.p3.16xlarge\",\n",
      "    \"distribution_hosts\": [\n",
      "        \"algo-1\",\n",
      "        \"algo-2\"\n",
      "    ],\n",
      "    \"distribution_instance_groups\": [\n",
      "        \"homogeneousCluster\"\n",
      "    ],\n",
      "    \"framework_module\": \"sagemaker_pytorch_container.training:main\",\n",
      "    \"hosts\": [\n",
      "        \"algo-1\",\n",
      "        \"algo-2\"\n",
      "    ],\n",
      "    \"hyperparameters\": {},\n",
      "    \"input_config_dir\": \"/opt/ml/input/config\",\n",
      "    \"input_data_config\": {},\n",
      "    \"input_dir\": \"/opt/ml/input\",\n",
      "    \"instance_groups\": [\n",
      "        \"homogeneousCluster\"\n",
      "    ],\n",
      "    \"instance_groups_dict\": {\n",
      "        \"homogeneousCluster\": {\n",
      "            \"instance_group_name\": \"homogeneousCluster\",\n",
      "            \"instance_type\": \"ml.p3.16xlarge\",\n",
      "            \"hosts\": [\n",
      "                \"algo-1\",\n",
      "                \"algo-2\"\n",
      "            ]\n",
      "        }\n",
      "    },\n",
      "    \"is_hetero\": false,\n",
      "    \"is_master\": true,\n",
      "    \"is_modelparallel_enabled\": null,\n",
      "    \"is_smddpmprun_installed\": true,\n",
      "    \"job_name\": \"pytorch-training-2023-02-02-20-42-59-071\",\n",
      "    \"log_level\": 20,\n",
      "    \"master_hostname\": \"algo-1\",\n",
      "    \"model_dir\": \"/opt/ml/model\",\n",
      "    \"module_dir\": \"s3://sagemaker-us-west-2-197237853439/pytorch-training-2023-02-02-20-42-59-071/source/sourcedir.tar.gz\",\n",
      "    \"module_name\": \"sagemaker_distributed_reading_service\",\n",
      "    \"network_interface_name\": \"eth0\",\n",
      "    \"num_cpus\": 64,\n",
      "    \"num_gpus\": 8,\n",
      "    \"num_neurons\": 0,\n",
      "    \"output_data_dir\": \"/opt/ml/output/data\",\n",
      "    \"output_dir\": \"/opt/ml/output\",\n",
      "    \"output_intermediate_dir\": \"/opt/ml/output/intermediate\",\n",
      "    \"resource_config\": {\n",
      "        \"current_host\": \"algo-1\",\n",
      "        \"current_instance_type\": \"ml.p3.16xlarge\",\n",
      "        \"current_group_name\": \"homogeneousCluster\",\n",
      "        \"hosts\": [\n",
      "            \"algo-1\",\n",
      "            \"algo-2\"\n",
      "        ],\n",
      "        \"instance_groups\": [\n",
      "            {\n",
      "                \"instance_group_name\": \"homogeneousCluster\",\n",
      "                \"instance_type\": \"ml.p3.16xlarge\",\n",
      "                \"hosts\": [\n",
      "                    \"algo-1\",\n",
      "                    \"algo-2\"\n",
      "                ]\n",
      "            }\n",
      "        ],\n",
      "        \"network_interface_name\": \"eth0\"\n",
      "    },\n",
      "    \"user_entry_point\": \"sagemaker_distributed_reading_service.py\"\n",
      "}\n",
      "Environment variables:\n",
      "SM_HOSTS=[\"algo-1\",\"algo-2\"]\n",
      "SM_NETWORK_INTERFACE_NAME=eth0\n",
      "SM_HPS={}\n",
      "SM_USER_ENTRY_POINT=sagemaker_distributed_reading_service.py\n",
      "SM_FRAMEWORK_PARAMS={\"sagemaker_distributed_dataparallel_custom_mpi_options\":\"\",\"sagemaker_distributed_dataparallel_enabled\":true,\"sagemaker_instance_type\":\"ml.p3.16xlarge\"}\n",
      "SM_RESOURCE_CONFIG={\"current_group_name\":\"homogeneousCluster\",\"current_host\":\"algo-1\",\"current_instance_type\":\"ml.p3.16xlarge\",\"hosts\":[\"algo-1\",\"algo-2\"],\"instance_groups\":[{\"hosts\":[\"algo-1\",\"algo-2\"],\"instance_group_name\":\"homogeneousCluster\",\"instance_type\":\"ml.p3.16xlarge\"}],\"network_interface_name\":\"eth0\"}\n",
      "SM_INPUT_DATA_CONFIG={}\n",
      "SM_OUTPUT_DATA_DIR=/opt/ml/output/data\n",
      "SM_CHANNELS=[]\n",
      "SM_CURRENT_HOST=algo-1\n",
      "SM_CURRENT_INSTANCE_TYPE=ml.p3.16xlarge\n",
      "SM_CURRENT_INSTANCE_GROUP=homogeneousCluster\n",
      "SM_CURRENT_INSTANCE_GROUP_HOSTS=[\"algo-1\",\"algo-2\"]\n",
      "SM_INSTANCE_GROUPS=[\"homogeneousCluster\"]\n",
      "SM_INSTANCE_GROUPS_DICT={\"homogeneousCluster\":{\"hosts\":[\"algo-1\",\"algo-2\"],\"instance_group_name\":\"homogeneousCluster\",\"instance_type\":\"ml.p3.16xlarge\"}}\n",
      "SM_DISTRIBUTION_INSTANCE_GROUPS=[\"homogeneousCluster\"]\n",
      "SM_IS_HETERO=false\n",
      "SM_MODULE_NAME=sagemaker_distributed_reading_service\n",
      "SM_LOG_LEVEL=20\n",
      "SM_FRAMEWORK_MODULE=sagemaker_pytorch_container.training:main\n",
      "SM_INPUT_DIR=/opt/ml/input\n",
      "SM_INPUT_CONFIG_DIR=/opt/ml/input/config\n",
      "SM_OUTPUT_DIR=/opt/ml/output\n",
      "SM_NUM_CPUS=64\n",
      "SM_NUM_GPUS=8\n",
      "SM_NUM_NEURONS=0\n",
      "SM_MODEL_DIR=/opt/ml/model\n",
      "SM_MODULE_DIR=s3://sagemaker-us-west-2-197237853439/pytorch-training-2023-02-02-20-42-59-071/source/sourcedir.tar.gz\n",
      "SM_TRAINING_ENV={\"additional_framework_parameters\":{\"sagemaker_distributed_dataparallel_custom_mpi_options\":\"\",\"sagemaker_distributed_dataparallel_enabled\":true,\"sagemaker_instance_type\":\"ml.p3.16xlarge\"},\"channel_input_dirs\":{},\"current_host\":\"algo-1\",\"current_instance_group\":\"homogeneousCluster\",\"current_instance_group_hosts\":[\"algo-1\",\"algo-2\"],\"current_instance_type\":\"ml.p3.16xlarge\",\"distribution_hosts\":[\"algo-1\",\"algo-2\"],\"distribution_instance_groups\":[\"homogeneousCluster\"],\"framework_module\":\"sagemaker_pytorch_container.training:main\",\"hosts\":[\"algo-1\",\"algo-2\"],\"hyperparameters\":{},\"input_config_dir\":\"/opt/ml/input/config\",\"input_data_config\":{},\"input_dir\":\"/opt/ml/input\",\"instance_groups\":[\"homogeneousCluster\"],\"instance_groups_dict\":{\"homogeneousCluster\":{\"hosts\":[\"algo-1\",\"algo-2\"],\"instance_group_name\":\"homogeneousCluster\",\"instance_type\":\"ml.p3.16xlarge\"}},\"is_hetero\":false,\"is_master\":true,\"is_modelparallel_enabled\":null,\"is_smddpmprun_installed\":true,\"job_name\":\"pytorch-training-2023-02-02-20-42-59-071\",\"log_level\":20,\"master_hostname\":\"algo-1\",\"model_dir\":\"/opt/ml/model\",\"module_dir\":\"s3://sagemaker-us-west-2-197237853439/pytorch-training-2023-02-02-20-42-59-071/source/sourcedir.tar.gz\",\"module_name\":\"sagemaker_distributed_reading_service\",\"network_interface_name\":\"eth0\",\"num_cpus\":64,\"num_gpus\":8,\"num_neurons\":0,\"output_data_dir\":\"/opt/ml/output/data\",\"output_dir\":\"/opt/ml/output\",\"output_intermediate_dir\":\"/opt/ml/output/intermediate\",\"resource_config\":{\"current_group_name\":\"homogeneousCluster\",\"current_host\":\"algo-1\",\"current_instance_type\":\"ml.p3.16xlarge\",\"hosts\":[\"algo-1\",\"algo-2\"],\"instance_groups\":[{\"hosts\":[\"algo-1\",\"algo-2\"],\"instance_group_name\":\"homogeneousCluster\",\"instance_type\":\"ml.p3.16xlarge\"}],\"network_interface_name\":\"eth0\"},\"user_entry_point\":\"sagemaker_distributed_reading_service.py\"}\n",
      "SM_USER_ARGS=[]\n",
      "SM_OUTPUT_INTERMEDIATE_DIR=/opt/ml/output/intermediate\n",
      "PYTHONPATH=/opt/ml/code:/opt/conda/bin:/opt/conda/lib/python39.zip:/opt/conda/lib/python3.9:/opt/conda/lib/python3.9/lib-dynload:/opt/conda/lib/python3.9/site-packages\n",
      "Invoking script with the following command:\n",
      "mpirun --host algo-1:8,algo-2:8 -np 16 --allow-run-as-root --tag-output --oversubscribe -mca btl_tcp_if_include eth0 -mca oob_tcp_if_include eth0 -mca plm_rsh_no_tree_spawn 1 -mca pml ob1 -mca btl ^openib -mca orte_abort_on_non_zero_status 1 -mca btl_vader_single_copy_mechanism none -mca plm_rsh_num_concurrent 2 -x NCCL_SOCKET_IFNAME=eth0 -x NCCL_DEBUG=INFO -x LD_LIBRARY_PATH -x PATH -x SMDATAPARALLEL_USE_HOMOGENEOUS=1 -x FI_PROVIDER=efa -x RDMAV_FORK_SAFE=1 -x LD_PRELOAD=/opt/conda/lib/python3.9/site-packages/gethostname.cpython-39-x86_64-linux-gnu.so -x SMDATAPARALLEL_SERVER_ADDR=algo-1 -x SMDATAPARALLEL_SERVER_PORT=7592 -x SAGEMAKER_INSTANCE_TYPE=ml.p3.16xlarge -x USE_SMDDP_COLLECTIVES=0 smddprun /opt/conda/bin/python3.9 -m mpi4py sagemaker_distributed_reading_service.py\n",
      "2023-02-02 20:49:25,892 root         INFO     Using NamedTuple = typing._NamedTuple instead.\n",
      "Warning: Permanently added 'algo-2,10.0.189.150' (ECDSA) to the list of known hosts.\n",
      "[1,mpirank:0,algo-1]<stderr>:curl: /opt/conda/lib/libcurl.so.4: no version information available (required by curl)\n",
      "[1,mpirank:0,algo-1]<stderr>:curl: /opt/conda/lib/libcurl.so.4: no version information available (required by curl)\n",
      "[1,mpirank:8,algo-2]<stderr>:curl: /opt/conda/lib/libcurl.so.4: no version information available (required by curl)\n",
      "[1,mpirank:8,algo-2]<stderr>:curl: /opt/conda/lib/libcurl.so.4: no version information available (required by curl)\n",
      "[1,mpirank:0,algo-1]<stderr>:curl: /opt/conda/lib/libcurl.so.4: no version information available (required by curl)\n",
      "[1,mpirank:8,algo-2]<stderr>:curl: /opt/conda/lib/libcurl.so.4: no version information available (required by curl)\n",
      "2023-02-02 20:49:27,100 sagemaker-training-toolkit INFO     Process[es]: [psutil.Process(pid=63, name='orted', status='sleeping', started='20:49:26')]\n",
      "2023-02-02 20:49:27,101 sagemaker-training-toolkit INFO     Orted process found [psutil.Process(pid=63, name='orted', status='sleeping', started='20:49:26')]\n",
      "2023-02-02 20:49:27,101 sagemaker-training-toolkit INFO     Waiting for orted process [psutil.Process(pid=63, name='orted', status='sleeping', started='20:49:26')]\n",
      "[1,mpirank:0,algo-1]<stdout>:DDP Mode\n",
      "[1,mpirank:7,algo-1]<stdout>:DDP Mode\n",
      "[1,mpirank:2,algo-1]<stdout>:DDP Mode\n",
      "[1,mpirank:4,algo-1]<stdout>:DDP Mode\n",
      "[1,mpirank:5,algo-1]<stdout>:DDP Mode\n",
      "[1,mpirank:1,algo-1]<stdout>:DDP Mode\n",
      "[1,mpirank:6,algo-1]<stdout>:DDP Mode\n",
      "[1,mpirank:3,algo-1]<stdout>:DDP Mode\n",
      "[1,mpirank:10,algo-2]<stdout>:DDP Mode\n",
      "[1,mpirank:15,algo-2]<stdout>:DDP Mode\n",
      "[1,mpirank:8,algo-2]<stdout>:DDP Mode\n",
      "[1,mpirank:13,algo-2]<stdout>:DDP Mode\n",
      "[1,mpirank:14,algo-2]<stdout>:DDP Mode\n",
      "[1,mpirank:11,algo-2]<stdout>:DDP Mode\n",
      "[1,mpirank:9,algo-2]<stdout>:DDP Mode\n",
      "[1,mpirank:12,algo-2]<stdout>:DDP Mode\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Bootstrap : Using eth0:10.0.150.213<0>\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO cudaDriverVersion 11070\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO cudaDriverVersion 11070\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO cudaDriverVersion 11070\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO cudaDriverVersion 11070\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO cudaDriverVersion 11070\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO cudaDriverVersion 11070\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO cudaDriverVersion 11070\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO cudaDriverVersion 11070\n",
      "[1,mpirank:0,algo-1]<stdout>:NCCL version 2.14.3+cuda11.7\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO cudaDriverVersion 11070\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO cudaDriverVersion 11070\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO cudaDriverVersion 11070\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO cudaDriverVersion 11070\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO cudaDriverVersion 11070\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO cudaDriverVersion 11070\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO cudaDriverVersion 11070\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO cudaDriverVersion 11070\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1\n",
      "[1,mpirank:0,algo-1]<stdout>:\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] find_ofi_provider:608 NCCL WARN NET/OFI Couldn't find any optimal provider\n",
      "[1,mpirank:0,algo-1]<stdout>:\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] ofi_init:1355 NCCL WARN NET/OFI aws-ofi-nccl initialization failed\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO NCCL_IB_DISABLE set by environment to 1.\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO NET/Socket : Using [0]eth0:10.0.150.213<0>\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Using network Socket\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Bootstrap : Using eth0:10.0.150.213<0>\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Bootstrap : Using eth0:10.0.150.213<0>\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1\n",
      "[1,mpirank:4,algo-1]<stdout>:\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] find_ofi_provider:608 NCCL WARN NET/OFI Couldn't find any optimal provider\n",
      "[1,mpirank:4,algo-1]<stdout>:\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] ofi_init:1355 NCCL WARN NET/OFI aws-ofi-nccl initialization failed\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO NCCL_IB_DISABLE set by environment to 1.\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO NET/Socket : Using [0]eth0:10.0.150.213<0>\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Using network Socket\n",
      "[1,mpirank:2,algo-1]<stdout>:\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] find_ofi_provider:608 NCCL WARN NET/OFI Couldn't find any optimal provider\n",
      "[1,mpirank:2,algo-1]<stdout>:\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] ofi_init:1355 NCCL WARN NET/OFI aws-ofi-nccl initialization failed\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO NCCL_IB_DISABLE set by environment to 1.\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO NET/Socket : Using [0]eth0:10.0.150.213<0>\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Using network Socket\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Bootstrap : Using eth0:10.0.189.150<0>\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Bootstrap : Using eth0:10.0.150.213<0>\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Bootstrap : Using eth0:10.0.189.150<0>\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Bootstrap : Using eth0:10.0.189.150<0>\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Bootstrap : Using eth0:10.0.150.213<0>\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1\n",
      "[1,mpirank:8,algo-2]<stdout>:\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] find_ofi_provider:608 NCCL WARN NET/OFI Couldn't find any optimal provider\n",
      "[1,mpirank:8,algo-2]<stdout>:\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] ofi_init:1355 NCCL WARN NET/OFI aws-ofi-nccl initialization failed\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO NCCL_IB_DISABLE set by environment to 1.\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO NET/Socket : Using [0]eth0:10.0.189.150<0>\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Using network Socket\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Bootstrap : Using eth0:10.0.150.213<0>\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Bootstrap : Using eth0:10.0.150.213<0>\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1\n",
      "[1,mpirank:3,algo-1]<stdout>:\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] find_ofi_provider:608 NCCL WARN NET/OFI Couldn't find any optimal provider\n",
      "[1,mpirank:3,algo-1]<stdout>:\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] ofi_init:1355 NCCL WARN NET/OFI aws-ofi-nccl initialization failed\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO NCCL_IB_DISABLE set by environment to 1.\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO NET/Socket : Using [0]eth0:10.0.150.213<0>\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Using network Socket\n",
      "[1,mpirank:9,algo-2]<stdout>:\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] find_ofi_provider:608 NCCL WARN NET/OFI Couldn't find any optimal provider\n",
      "[1,mpirank:9,algo-2]<stdout>:\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] ofi_init:1355 NCCL WARN NET/OFI aws-ofi-nccl initialization failed\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO NCCL_IB_DISABLE set by environment to 1.\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO NET/Socket : Using [0]eth0:10.0.189.150<0>\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Using network Socket\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Bootstrap : Using eth0:10.0.189.150<0>\n",
      "[1,mpirank:10,algo-2]<stdout>:\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] find_ofi_provider:608 NCCL WARN NET/OFI Couldn't find any optimal provider\n",
      "[1,mpirank:10,algo-2]<stdout>:\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] ofi_init:1355 NCCL WARN NET/OFI aws-ofi-nccl initialization failed\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO NCCL_IB_DISABLE set by environment to 1.\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO NET/Socket : Using [0]eth0:10.0.189.150<0>\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Using network Socket\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).\n",
      "[1,mpirank:7,algo-1]<stdout>:\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] find_ofi_provider:608 NCCL WARN NET/OFI Couldn't find any optimal provider\n",
      "[1,mpirank:7,algo-1]<stdout>:\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] ofi_init:1355 NCCL WARN NET/OFI aws-ofi-nccl initialization failed\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO NCCL_IB_DISABLE set by environment to 1.\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO NET/Socket : Using [0]eth0:10.0.150.213<0>\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Using network Socket\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Bootstrap : Using eth0:10.0.189.150<0>\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1\n",
      "[1,mpirank:1,algo-1]<stdout>:\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] find_ofi_provider:608 NCCL WARN NET/OFI Couldn't find any optimal provider\n",
      "[1,mpirank:1,algo-1]<stdout>:\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] ofi_init:1355 NCCL WARN NET/OFI aws-ofi-nccl initialization failed\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO NCCL_IB_DISABLE set by environment to 1.\n",
      "[1,mpirank:5,algo-1]<stdout>:\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] find_ofi_provider:608 NCCL WARN NET/OFI Couldn't find any optimal provider\n",
      "[1,mpirank:5,algo-1]<stdout>:\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] ofi_init:1355 NCCL WARN NET/OFI aws-ofi-nccl initialization failed\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO NCCL_IB_DISABLE set by environment to 1.\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO NET/Socket : Using [0]eth0:10.0.150.213<0>\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Using network Socket\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO NET/Socket : Using [0]eth0:10.0.150.213<0>\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Using network Socket\n",
      "[1,mpirank:14,algo-2]<stdout>:\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] find_ofi_provider:608 NCCL WARN NET/OFI Couldn't find any optimal provider\n",
      "[1,mpirank:14,algo-2]<stdout>:\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] ofi_init:1355 NCCL WARN NET/OFI aws-ofi-nccl initialization failed\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO NCCL_IB_DISABLE set by environment to 1.\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO NET/Socket : Using [0]eth0:10.0.189.150<0>\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Using network Socket\n",
      "[1,mpirank:15,algo-2]<stdout>:\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] find_ofi_provider:608 NCCL WARN NET/OFI Couldn't find any optimal provider\n",
      "[1,mpirank:15,algo-2]<stdout>:\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] ofi_init:1355 NCCL WARN NET/OFI aws-ofi-nccl initialization failed\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO NCCL_IB_DISABLE set by environment to 1.\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO NET/Socket : Using [0]eth0:10.0.189.150<0>\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Using network Socket\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Bootstrap : Using eth0:10.0.150.213<0>\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1\n",
      "[1,mpirank:6,algo-1]<stdout>:\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] find_ofi_provider:608 NCCL WARN NET/OFI Couldn't find any optimal provider\n",
      "[1,mpirank:6,algo-1]<stdout>:\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] ofi_init:1355 NCCL WARN NET/OFI aws-ofi-nccl initialization failed\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO NCCL_IB_DISABLE set by environment to 1.\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO NET/Socket : Using [0]eth0:10.0.150.213<0>\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Using network Socket\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Bootstrap : Using eth0:10.0.189.150<0>\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1\n",
      "[1,mpirank:12,algo-2]<stdout>:\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] find_ofi_provider:608 NCCL WARN NET/OFI Couldn't find any optimal provider\n",
      "[1,mpirank:12,algo-2]<stdout>:\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] ofi_init:1355 NCCL WARN NET/OFI aws-ofi-nccl initialization failed\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO NCCL_IB_DISABLE set by environment to 1.\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO NET/Socket : Using [0]eth0:10.0.189.150<0>\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Using network Socket\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Bootstrap : Using eth0:10.0.189.150<0>\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1\n",
      "[1,mpirank:13,algo-2]<stdout>:\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] find_ofi_provider:608 NCCL WARN NET/OFI Couldn't find any optimal provider\n",
      "[1,mpirank:13,algo-2]<stdout>:\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] ofi_init:1355 NCCL WARN NET/OFI aws-ofi-nccl initialization failed\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO NCCL_IB_DISABLE set by environment to 1.\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO NET/Socket : Using [0]eth0:10.0.189.150<0>\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Using network Socket\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Bootstrap : Using eth0:10.0.189.150<0>\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin_v6 symbol.\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO NET/Plugin: Failed to find ncclCollNetPlugin symbol (v4 or v5).\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO NET/OFI Using aws-ofi-nccl 1.4.0aws\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO NET/OFI Setting FI_EFA_FORK_SAFE environment variable to 1\n",
      "[1,mpirank:11,algo-2]<stdout>:\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] find_ofi_provider:608 NCCL WARN NET/OFI Couldn't find any optimal provider\n",
      "[1,mpirank:11,algo-2]<stdout>:\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] ofi_init:1355 NCCL WARN NET/OFI aws-ofi-nccl initialization failed\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO NCCL_IB_DISABLE set by environment to 1.\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO NET/Socket : Using [0]eth0:10.0.189.150<0>\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Using network Socket\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Trees [0] 1/-1/-1->2->3 [1] 1/-1/-1->2->3\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Trees [0] 4/-1/-1->7->6 [1] 4/-1/-1->7->6\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Trees [0] 5/-1/-1->1->2 [1] 5/-1/-1->1->2\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Trees [0] 6/-1/-1->5->1 [1] 6/-1/-1->5->1\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Trees [0] -1/-1/-1->4->7 [1] -1/-1/-1->4->7\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 00/02 :    0   3   2   1   5   6   7   4   8  11  10   9  13  14  15  12\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 01/02 :    0   3   2   1   5   6   7   4   8  11  10   9  13  14  15  12\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Trees [0] 3/8/-1->0->-1 [1] 3/-1/-1->0->8\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Trees [0] 2/-1/-1->3->0 [1] 2/-1/-1->3->0\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Trees [0] 7/-1/-1->6->5 [1] 7/-1/-1->6->5\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Trees [0] 13/-1/-1->9->10 [1] 13/-1/-1->9->10\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Trees [0] 9/-1/-1->10->11 [1] 9/-1/-1->10->11\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Trees [0] 12/-1/-1->15->14 [1] 12/-1/-1->15->14\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Trees [0] 14/-1/-1->13->9 [1] 14/-1/-1->13->9\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Trees [0] 11/-1/-1->8->0 [1] 11/0/-1->8->-1\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Trees [0] 15/-1/-1->14->13 [1] 15/-1/-1->14->13\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Trees [0] -1/-1/-1->12->15 [1] -1/-1/-1->12->15\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Trees [0] 10/-1/-1->11->8 [1] 10/-1/-1->11->8\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 00/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 00/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 00/0 : 0[170] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 00/0 : 13[1c0] -> 14[1d0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 00/0 : 9[180] -> 13[1c0] via P2P/IPC\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 00/0 : 8[170] -> 11[1a0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 01/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 01/0 : 0[170] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 01/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 01/0 : 13[1c0] -> 14[1d0] via P2P/IPC\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 01/0 : 8[170] -> 11[1a0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 01/0 : 9[180] -> 13[1c0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 00/0 : 4[1b0] -> 8[170] [send] via NET/Socket/0\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 00/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 00/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 00/0 : 12[1b0] -> 0[170] [send] via NET/Socket/0\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 00/0 : 14[1d0] -> 15[1e0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 00/0 : 10[190] -> 9[180] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 01/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 01/0 : 14[1d0] -> 15[1e0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 01/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 01/0 : 10[190] -> 9[180] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 00/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 00/0 : 11[1a0] -> 10[190] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 01/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 01/0 : 11[1a0] -> 10[190] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 00/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 00/0 : 15[1e0] -> 12[1b0] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 01/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Connected all rings\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Connected all rings\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 01/0 : 15[1e0] -> 12[1b0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Connected all rings\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Connected all rings\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Connected all rings\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 00/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 00/0 : 14[1d0] -> 13[1c0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 00/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Connected all rings\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 00/0 : 13[1c0] -> 9[180] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 01/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 01/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 01/0 : 14[1d0] -> 13[1c0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 01/0 : 13[1c0] -> 9[180] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Connected all rings\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 00/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Connected all rings\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 00/0 : 9[180] -> 10[190] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 01/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Connected all rings\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 01/0 : 9[180] -> 10[190] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Connected all rings\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 00/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 01/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 00/0 : 10[190] -> 11[1a0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 01/0 : 10[190] -> 11[1a0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 00/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 00/0 : 11[1a0] -> 8[170] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Connected all trees\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO 2 coll channels, 2 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 00/1 : 2[190] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 01/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 01/0 : 11[1a0] -> 8[170] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Connected all trees\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO 2 coll channels, 2 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 00/1 : 10[190] -> 12[1b0] via P2P/indirect/8[170]\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Connected all trees\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO 2 coll channels, 2 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 00/1 : 1[180] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Connected all trees\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO 2 coll channels, 2 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 00/1 : 9[180] -> 12[1b0] via P2P/indirect/8[170]\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Connected all trees\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO 2 coll channels, 2 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Connected all trees\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO 2 coll channels, 2 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 01/1 : 2[190] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 01/1 : 10[190] -> 12[1b0] via P2P/indirect/8[170]\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 01/1 : 1[180] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 01/1 : 9[180] -> 12[1b0] via P2P/indirect/8[170]\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 01/0 : 12[1b0] -> 0[170] [send] via NET/Socket/0\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:782 [0] NCCL INFO NET/Socket: Using 2 threads and 8 sockets per thread\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 00/0 : 4[1b0] -> 8[170] [receive] via NET/Socket/0\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 01/0 : 4[1b0] -> 8[170] [send] via NET/Socket/0\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:846 [0] NCCL INFO NET/Socket: Using 2 threads and 8 sockets per thread\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 00/0 : 12[1b0] -> 0[170] [receive] via NET/Socket/0\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:782 [0] NCCL INFO NET/Socket: Using 2 threads and 8 sockets per thread\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 01/0 : 4[1b0] -> 8[170] [receive] via NET/Socket/0\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:846 [0] NCCL INFO NET/Socket: Using 2 threads and 8 sockets per thread\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 01/0 : 12[1b0] -> 0[170] [receive] via NET/Socket/0\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Connected all rings\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Connected all rings\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Connected all rings\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 00/0 : 12[1b0] -> 15[1e0] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Connected all rings\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 01/0 : 12[1b0] -> 15[1e0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Connected all rings\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 00/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Connected all rings\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 01/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Connected all trees\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO 2 coll channels, 2 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 00/0 : 15[1e0] -> 14[1d0] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 01/0 : 15[1e0] -> 14[1d0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Connected all trees\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO 2 coll channels, 2 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 00/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 01/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Connected all trees\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO 2 coll channels, 2 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Connected all trees\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO 2 coll channels, 2 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Connected all trees\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO 2 coll channels, 2 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Connected all trees\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO 2 coll channels, 2 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:782 [0] NCCL INFO NET/Socket: Using 2 threads and 8 sockets per thread\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 00/0 : 0[170] -> 8[170] [receive] via NET/Socket/0\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:846 [0] NCCL INFO NET/Socket: Using 2 threads and 8 sockets per thread\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 00/0 : 8[170] -> 0[170] [receive] via NET/Socket/0\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:782 [0] NCCL INFO NET/Socket: Using 2 threads and 8 sockets per thread\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 01/0 : 0[170] -> 8[170] [receive] via NET/Socket/0\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:846 [0] NCCL INFO NET/Socket: Using 2 threads and 8 sockets per thread\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 01/0 : 8[170] -> 0[170] [receive] via NET/Socket/0\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 00/0 : 8[170] -> 0[170] [send] via NET/Socket/0\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 00/0 : 0[170] -> 8[170] [send] via NET/Socket/0\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 01/0 : 8[170] -> 0[170] [send] via NET/Socket/0\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 01/0 : 0[170] -> 8[170] [send] via NET/Socket/0\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Connected all trees\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Connected all trees\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO 2 coll channels, 2 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 00/1 : 8[170] -> 13[1c0] via P2P/indirect/9[180]\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO 2 coll channels, 2 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 00/1 : 0[170] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Connected all trees\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO 2 coll channels, 2 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 00/1 : 11[1a0] -> 12[1b0] via P2P/indirect/8[170]\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Connected all trees\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO 2 coll channels, 2 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 00/1 : 3[1a0] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 01/1 : 11[1a0] -> 12[1b0] via P2P/indirect/8[170]\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 01/1 : 8[170] -> 13[1c0] via P2P/indirect/9[180]\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 01/1 : 3[1a0] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 01/1 : 0[170] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 00/1 : 11[1a0] -> 13[1c0] via P2P/indirect/9[180]\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 00/1 : 3[1a0] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 01/1 : 11[1a0] -> 13[1c0] via P2P/indirect/9[180]\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 01/1 : 3[1a0] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 00/1 : 11[1a0] -> 14[1d0] via P2P/indirect/15[1e0]\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 00/1 : 10[190] -> 13[1c0] via P2P/indirect/9[180]\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 00/1 : 3[1a0] -> 6[1d0] via P2P/indirect/7[1e0]\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 00/1 : 2[190] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 01/1 : 11[1a0] -> 14[1d0] via P2P/indirect/15[1e0]\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 01/1 : 10[190] -> 13[1c0] via P2P/indirect/9[180]\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 01/1 : 3[1a0] -> 6[1d0] via P2P/indirect/7[1e0]\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 01/1 : 2[190] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 00/1 : 12[1b0] -> 9[180] via P2P/indirect/13[1c0]\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 00/1 : 9[180] -> 14[1d0] via P2P/indirect/13[1c0]\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 00/1 : 4[1b0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 00/1 : 1[180] -> 6[1d0] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 00/1 : 10[190] -> 15[1e0] via P2P/indirect/14[1d0]\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 00/1 : 2[190] -> 7[1e0] via P2P/indirect/6[1d0]\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 01/1 : 12[1b0] -> 9[180] via P2P/indirect/13[1c0]\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 01/1 : 4[1b0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 01/1 : 9[180] -> 14[1d0] via P2P/indirect/13[1c0]\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 01/1 : 1[180] -> 6[1d0] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 01/1 : 10[190] -> 15[1e0] via P2P/indirect/14[1d0]\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 01/1 : 2[190] -> 7[1e0] via P2P/indirect/6[1d0]\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 00/1 : 9[180] -> 15[1e0] via P2P/indirect/11[1a0]\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 00/1 : 1[180] -> 7[1e0] via P2P/indirect/3[1a0]\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 00/1 : 13[1c0] -> 8[170] via P2P/indirect/12[1b0]\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 00/1 : 8[170] -> 14[1d0] via P2P/indirect/12[1b0]\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 00/1 : 5[1c0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 00/1 : 0[170] -> 6[1d0] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 01/1 : 13[1c0] -> 8[170] via P2P/indirect/12[1b0]\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 01/1 : 9[180] -> 15[1e0] via P2P/indirect/11[1a0]\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 01/1 : 5[1c0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 01/1 : 1[180] -> 7[1e0] via P2P/indirect/3[1a0]\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 01/1 : 8[170] -> 14[1d0] via P2P/indirect/12[1b0]\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 01/1 : 0[170] -> 6[1d0] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 00/1 : 14[1d0] -> 8[170] via P2P/indirect/12[1b0]\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 00/1 : 8[170] -> 15[1e0] via P2P/indirect/12[1b0]\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 00/1 : 6[1d0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 00/1 : 0[170] -> 7[1e0] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 01/1 : 14[1d0] -> 8[170] via P2P/indirect/12[1b0]\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 01/1 : 6[1d0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 01/1 : 8[170] -> 15[1e0] via P2P/indirect/12[1b0]\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 01/1 : 0[170] -> 7[1e0] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 00/1 : 7[1e0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 00/1 : 15[1e0] -> 8[170] via P2P/indirect/12[1b0]\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 01/1 : 7[1e0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 01/1 : 15[1e0] -> 8[170] via P2P/indirect/12[1b0]\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 00/1 : 7[1e0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 00/1 : 15[1e0] -> 9[180] via P2P/indirect/13[1c0]\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 01/1 : 7[1e0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 01/1 : 15[1e0] -> 9[180] via P2P/indirect/13[1c0]\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 00/1 : 6[1d0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 00/1 : 14[1d0] -> 9[180] via P2P/indirect/13[1c0]\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 00/1 : 7[1e0] -> 2[190] via P2P/indirect/3[1a0]\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 00/1 : 15[1e0] -> 10[190] via P2P/indirect/11[1a0]\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 01/1 : 6[1d0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 01/1 : 7[1e0] -> 2[190] via P2P/indirect/3[1a0]\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 01/1 : 15[1e0] -> 10[190] via P2P/indirect/11[1a0]\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 01/1 : 14[1d0] -> 9[180] via P2P/indirect/13[1c0]\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 00/1 : 5[1c0] -> 2[190] via P2P/indirect/1[180]\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 00/1 : 13[1c0] -> 10[190] via P2P/indirect/9[180]\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 01/1 : 5[1c0] -> 2[190] via P2P/indirect/1[180]\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 01/1 : 13[1c0] -> 10[190] via P2P/indirect/9[180]\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 00/1 : 6[1d0] -> 3[1a0] via P2P/indirect/2[190]\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 00/1 : 14[1d0] -> 11[1a0] via P2P/indirect/10[190]\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 01/1 : 6[1d0] -> 3[1a0] via P2P/indirect/2[190]\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 01/1 : 14[1d0] -> 11[1a0] via P2P/indirect/10[190]\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 00/1 : 5[1c0] -> 3[1a0] via P2P/indirect/1[180]\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 00/1 : 13[1c0] -> 11[1a0] via P2P/indirect/9[180]\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 00/1 : 12[1b0] -> 10[190] via P2P/indirect/14[1d0]\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 01/1 : 5[1c0] -> 3[1a0] via P2P/indirect/1[180]\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 00/1 : 4[1b0] -> 2[190] via P2P/indirect/6[1d0]\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 01/1 : 13[1c0] -> 11[1a0] via P2P/indirect/9[180]\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 01/1 : 12[1b0] -> 10[190] via P2P/indirect/14[1d0]\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 00/1 : 12[1b0] -> 11[1a0] via P2P/indirect/8[170]\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 01/1 : 4[1b0] -> 2[190] via P2P/indirect/6[1d0]\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 01/1 : 12[1b0] -> 11[1a0] via P2P/indirect/8[170]\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 00/1 : 4[1b0] -> 3[1a0] via P2P/indirect/0[170]\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 01/1 : 4[1b0] -> 3[1a0] via P2P/indirect/0[170]\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO comm 0x5596305c56e0 rank 15 nranks 16 cudaDev 7 busId 1e0 - Init COMPLETE\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO comm 0x55f35245f300 rank 9 nranks 16 cudaDev 1 busId 180 - Init COMPLETE\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO comm 0x55e18ae73bf0 rank 11 nranks 16 cudaDev 3 busId 1a0 - Init COMPLETE\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO comm 0x55e6c5646330 rank 10 nranks 16 cudaDev 2 busId 190 - Init COMPLETE\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO comm 0x55e37962f5b0 rank 14 nranks 16 cudaDev 6 busId 1d0 - Init COMPLETE\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO comm 0x55a10ec93d90 rank 12 nranks 16 cudaDev 4 busId 1b0 - Init COMPLETE\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO comm 0x55bb92b7c1c0 rank 4 nranks 16 cudaDev 4 busId 1b0 - Init COMPLETE\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO comm 0x55a8dd4e92a0 rank 2 nranks 16 cudaDev 2 busId 190 - Init COMPLETE\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO comm 0x555de32f4a20 rank 3 nranks 16 cudaDev 3 busId 1a0 - Init COMPLETE\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO comm 0x55a150dbc9a0 rank 6 nranks 16 cudaDev 6 busId 1d0 - Init COMPLETE\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO comm 0x55850f585e40 rank 0 nranks 16 cudaDev 0 busId 170 - Init COMPLETE\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO comm 0x5588bf1d0b30 rank 5 nranks 16 cudaDev 5 busId 1c0 - Init COMPLETE\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Using network Socket\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Using network Socket\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Using network Socket\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Using network Socket\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Using network Socket\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Using network Socket\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO comm 0x5626ddce3bc0 rank 7 nranks 16 cudaDev 7 busId 1e0 - Init COMPLETE\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Using network Socket\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO comm 0x55a2b65b99c0 rank 1 nranks 16 cudaDev 1 busId 180 - Init COMPLETE\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Using network Socket\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO comm 0x55ec5e6d4700 rank 13 nranks 16 cudaDev 5 busId 1c0 - Init COMPLETE\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO comm 0x55c1988e94c0 rank 8 nranks 16 cudaDev 0 busId 170 - Init COMPLETE\n",
      "[1,mpirank:8,algo-2]<stdout>:NCCL version 2.14.3+cuda11.7\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Using network Socket\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Using network Socket\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Using network Socket\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Using network Socket\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Using network Socket\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Using network Socket\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Using network Socket\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Using network Socket\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Trees [0] 4/-1/-1->7->6 [1] 4/-1/-1->7->6 [2] 6/-1/-1->7->4 [3] 6/-1/-1->7->4 [4] 5/-1/-1->7->3 [5] 3/-1/-1->7->5 [6] 4/-1/-1->7->6 [7] 4/-1/-1->7->6 [8] 6/-1/-1->7->4 [9] 6/-1/-1->7->4 [10] 5/-1/-1->7->3 [11] 3/-1/-1->7->5\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Trees [0] 7/-1/-1->6->5 [1] 7/-1/-1->6->5 [2] 5/-1/-1->6->7 [3] 5/-1/-1->6->7 [4] 2/-1/-1->6->4 [5] 4/-1/-1->6->2 [6] 7/-1/-1->6->5 [7] 7/-1/-1->6->5 [8] 5/-1/-1->6->7 [9] 5/-1/-1->6->7 [10] 2/-1/-1->6->4 [11] 4/-1/-1->6->2\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 00/12 :    0   3   2   1   5   6   7   4\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 01/12 :    0   3   2   1   5   6   7   4\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 02/12 :    0   4   7   6   5   1   2   3\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 03/12 :    0   4   7   6   5   1   2   3\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 04/12 :    0   1   3   7   5   4   6   2\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 05/12 :    0   2   6   4   5   7   3   1\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 06/12 :    0   3   2   1   5   6   7   4\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 07/12 :    0   3   2   1   5   6   7   4\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 08/12 :    0   4   7   6   5   1   2   3\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 09/12 :    0   4   7   6   5   1   2   3\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 10/12 :    0   1   3   7   5   4   6   2\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 11/12 :    0   2   6   4   5   7   3   1\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Trees [0] 3/-1/-1->0->-1 [1] 3/-1/-1->0->-1 [2] 4/-1/-1->0->-1 [3] 4/-1/-1->0->-1 [4] 1/-1/-1->0->-1 [5] 2/-1/-1->0->-1 [6] 3/-1/-1->0->-1 [7] 3/-1/-1->0->-1 [8] 4/-1/-1->0->-1 [9] 4/-1/-1->0->-1 [10] 1/-1/-1->0->-1 [11] 2/-1/-1->0->-1\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Trees [0] -1/-1/-1->4->7 [1] -1/-1/-1->4->7 [2] 7/-1/-1->4->0 [3] 7/-1/-1->4->0 [4] 6/-1/-1->4->5 [5] 5/-1/-1->4->6 [6] -1/-1/-1->4->7 [7] -1/-1/-1->4->7 [8] 7/-1/-1->4->0 [9] 7/-1/-1->4->0 [10] 6/-1/-1->4->5 [11] 5/-1/-1->4->6\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Trees [0] 1/-1/-1->2->3 [1] 1/-1/-1->2->3 [2] 3/-1/-1->2->1 [3] 3/-1/-1->2->1 [4] -1/-1/-1->2->6 [5] 6/-1/-1->2->0 [6] 1/-1/-1->2->3 [7] 1/-1/-1->2->3 [8] 3/-1/-1->2->1 [9] 3/-1/-1->2->1 [10] -1/-1/-1->2->6 [11] 6/-1/-1->2->0\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Trees [0] 2/-1/-1->3->0 [1] 2/-1/-1->3->0 [2] -1/-1/-1->3->2 [3] -1/-1/-1->3->2 [4] 7/-1/-1->3->1 [5] 1/-1/-1->3->7 [6] 2/-1/-1->3->0 [7] 2/-1/-1->3->0 [8] -1/-1/-1->3->2 [9] -1/-1/-1->3->2 [10] 7/-1/-1->3->1 [11] 1/-1/-1->3->7\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Trees [0] 6/-1/-1->5->1 [1] 6/-1/-1->5->1 [2] 1/-1/-1->5->6 [3] 1/-1/-1->5->6 [4] 4/-1/-1->5->7 [5] 7/-1/-1->5->4 [6] 6/-1/-1->5->1 [7] 6/-1/-1->5->1 [8] 1/-1/-1->5->6 [9] 1/-1/-1->5->6 [10] 4/-1/-1->5->7 [11] 7/-1/-1->5->4\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Trees [0] 5/-1/-1->1->2 [1] 5/-1/-1->1->2 [2] 2/-1/-1->1->5 [3] 2/-1/-1->1->5 [4] 3/-1/-1->1->0 [5] -1/-1/-1->1->3 [6] 5/-1/-1->1->2 [7] 5/-1/-1->1->2 [8] 2/-1/-1->1->5 [9] 2/-1/-1->1->5 [10] 3/-1/-1->1->0 [11] -1/-1/-1->1->3\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 05/0 : 4[1b0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 04/0 : 0[170] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 10/0 : 0[170] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 11/0 : 4[1b0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 00/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 02/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 01/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 03/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 02/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 06/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 00/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 08/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 03/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 07/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 01/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 09/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 08/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 06/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 09/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 07/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 04/0 : 4[1b0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 05/0 : 0[170] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 10/0 : 4[1b0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 11/0 : 0[170] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Trees [0] 2/-1/-1->3->0 [1] 2/-1/-1->3->0 [2] -1/-1/-1->3->2 [3] -1/-1/-1->3->2 [4] 7/-1/-1->3->1 [5] 1/-1/-1->3->7 [6] 2/-1/-1->3->0 [7] 2/-1/-1->3->0 [8] -1/-1/-1->3->2 [9] -1/-1/-1->3->2 [10] 7/-1/-1->3->1 [11] 1/-1/-1->3->7\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Trees [0] 5/-1/-1->1->2 [1] 5/-1/-1->1->2 [2] 2/-1/-1->1->5 [3] 2/-1/-1->1->5 [4] 3/-1/-1->1->0 [5] -1/-1/-1->1->3 [6] 5/-1/-1->1->2 [7] 5/-1/-1->1->2 [8] 2/-1/-1->1->5 [9] 2/-1/-1->1->5 [10] 3/-1/-1->1->0 [11] -1/-1/-1->1->3\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Trees [0] 6/-1/-1->5->1 [1] 6/-1/-1->5->1 [2] 1/-1/-1->5->6 [3] 1/-1/-1->5->6 [4] 4/-1/-1->5->7 [5] 7/-1/-1->5->4 [6] 6/-1/-1->5->1 [7] 6/-1/-1->5->1 [8] 1/-1/-1->5->6 [9] 1/-1/-1->5->6 [10] 4/-1/-1->5->7 [11] 7/-1/-1->5->4\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Trees [0] 4/-1/-1->7->6 [1] 4/-1/-1->7->6 [2] 6/-1/-1->7->4 [3] 6/-1/-1->7->4 [4] 5/-1/-1->7->3 [5] 3/-1/-1->7->5 [6] 4/-1/-1->7->6 [7] 4/-1/-1->7->6 [8] 6/-1/-1->7->4 [9] 6/-1/-1->7->4 [10] 5/-1/-1->7->3 [11] 3/-1/-1->7->5\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Trees [0] -1/-1/-1->4->7 [1] -1/-1/-1->4->7 [2] 7/-1/-1->4->0 [3] 7/-1/-1->4->0 [4] 6/-1/-1->4->5 [5] 5/-1/-1->4->6 [6] -1/-1/-1->4->7 [7] -1/-1/-1->4->7 [8] 7/-1/-1->4->0 [9] 7/-1/-1->4->0 [10] 6/-1/-1->4->5 [11] 5/-1/-1->4->6\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Trees [0] 1/-1/-1->2->3 [1] 1/-1/-1->2->3 [2] 3/-1/-1->2->1 [3] 3/-1/-1->2->1 [4] -1/-1/-1->2->6 [5] 6/-1/-1->2->0 [6] 1/-1/-1->2->3 [7] 1/-1/-1->2->3 [8] 3/-1/-1->2->1 [9] 3/-1/-1->2->1 [10] -1/-1/-1->2->6 [11] 6/-1/-1->2->0\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Trees [0] 7/-1/-1->6->5 [1] 7/-1/-1->6->5 [2] 5/-1/-1->6->7 [3] 5/-1/-1->6->7 [4] 2/-1/-1->6->4 [5] 4/-1/-1->6->2 [6] 7/-1/-1->6->5 [7] 7/-1/-1->6->5 [8] 5/-1/-1->6->7 [9] 5/-1/-1->6->7 [10] 2/-1/-1->6->4 [11] 4/-1/-1->6->2\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 00/12 :    0   3   2   1   5   6   7   4\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 01/12 :    0   3   2   1   5   6   7   4\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 02/12 :    0   4   7   6   5   1   2   3\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 03/12 :    0   4   7   6   5   1   2   3\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 04/12 :    0   1   3   7   5   4   6   2\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 05/12 :    0   2   6   4   5   7   3   1\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 06/12 :    0   3   2   1   5   6   7   4\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 07/12 :    0   3   2   1   5   6   7   4\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 08/12 :    0   4   7   6   5   1   2   3\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 09/12 :    0   4   7   6   5   1   2   3\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 10/12 :    0   1   3   7   5   4   6   2\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 11/12 :    0   2   6   4   5   7   3   1\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Trees [0] 3/-1/-1->0->-1 [1] 3/-1/-1->0->-1 [2] 4/-1/-1->0->-1 [3] 4/-1/-1->0->-1 [4] 1/-1/-1->0->-1 [5] 2/-1/-1->0->-1 [6] 3/-1/-1->0->-1 [7] 3/-1/-1->0->-1 [8] 4/-1/-1->0->-1 [9] 4/-1/-1->0->-1 [10] 1/-1/-1->0->-1 [11] 2/-1/-1->0->-1\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 05/0 : 4[1b0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 04/0 : 0[170] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 11/0 : 4[1b0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 04/0 : 1[180] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 10/0 : 0[170] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 05/0 : 5[1c0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 10/0 : 1[180] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 02/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 00/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 11/0 : 5[1c0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 03/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 01/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 02/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 08/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 06/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 00/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 03/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 07/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 01/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 09/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 08/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 06/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 09/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 07/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 05/0 : 0[170] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 04/0 : 4[1b0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 00/0 : 0[170] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 11/0 : 0[170] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 02/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 10/0 : 4[1b0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 01/0 : 0[170] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 03/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 05/0 : 2[190] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 06/0 : 0[170] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 04/0 : 6[1d0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 08/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 00/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 11/0 : 2[190] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 07/0 : 0[170] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 02/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 10/0 : 6[1d0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 09/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 01/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 04/0 : 1[180] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 03/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 06/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 05/0 : 5[1c0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 10/0 : 1[180] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 08/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 11/0 : 5[1c0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 07/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 09/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 04/0 : 2[190] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 05/0 : 6[1d0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 00/0 : 0[170] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 10/0 : 2[190] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 11/0 : 6[1d0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 04/0 : 3[1a0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 02/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 01/0 : 0[170] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 05/0 : 7[1e0] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 10/0 : 3[1a0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 03/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 05/0 : 2[190] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 06/0 : 0[170] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 11/0 : 7[1e0] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 02/0 : 0[170] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 00/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 08/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 11/0 : 2[190] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 04/0 : 6[1d0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 07/0 : 0[170] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 02/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 00/0 : 4[1b0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 09/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 01/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 10/0 : 6[1d0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 03/0 : 0[170] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 03/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 06/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 01/0 : 4[1b0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 08/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 08/0 : 0[170] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 07/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 06/0 : 4[1b0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 02/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 09/0 : 0[170] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 00/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 09/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 07/0 : 4[1b0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 03/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 01/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 08/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 06/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 09/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 07/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 04/0 : 2[190] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 05/0 : 6[1d0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 10/0 : 2[190] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 04/0 : 3[1a0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 11/0 : 6[1d0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 05/0 : 7[1e0] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 10/0 : 3[1a0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 11/0 : 7[1e0] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 05/0 : 3[1a0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 02/0 : 0[170] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 04/0 : 7[1e0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 00/0 : 4[1b0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 11/0 : 3[1a0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 03/0 : 0[170] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 10/0 : 7[1e0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 01/0 : 4[1b0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 08/0 : 0[170] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 06/0 : 4[1b0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 09/0 : 0[170] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 02/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 00/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 00/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 07/0 : 4[1b0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 03/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 01/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 08/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 06/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 02/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 01/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 09/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 07/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 03/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 06/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 08/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 07/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 09/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 00/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 05/0 : 1[180] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 02/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 04/0 : 5[1c0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 01/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 03/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 11/0 : 1[180] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 10/0 : 5[1c0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 06/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 08/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 07/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 05/0 : 3[1a0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 04/0 : 7[1e0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 09/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Connected all rings\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 11/0 : 3[1a0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 10/0 : 7[1e0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Connected all rings\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 04/0 : 4[1b0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 10/0 : 4[1b0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 00/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 02/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Connected all rings\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 01/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 03/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Connected all rings\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 06/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 08/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Connected all rings\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 00/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Connected all rings\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 07/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 09/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 00/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 01/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Connected all rings\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 02/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 05/0 : 1[180] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 04/0 : 5[1c0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 01/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Connected all rings\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 03/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 11/0 : 1[180] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 10/0 : 5[1c0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 06/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 06/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 02/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 08/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 07/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 07/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 09/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 03/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Connected all rings\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 04/0 : 4[1b0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Connected all rings\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 08/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 02/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 00/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 09/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 01/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 03/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 10/0 : 4[1b0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 06/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 08/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 05/0 : 4[1b0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 09/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 07/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 11/0 : 4[1b0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Connected all rings\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Connected all rings\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Connected all rings\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 00/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 05/0 : 1[180] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Connected all rings\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 01/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 11/0 : 1[180] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 04/0 : 5[1c0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Connected all rings\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Connected all rings\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 06/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 10/0 : 5[1c0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 02/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 07/0 : 1[180] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 03/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 08/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 00/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 09/0 : 5[1c0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 02/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 01/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 03/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 04/0 : 2[190] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 06/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 08/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 05/0 : 4[1b0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 07/0 : 2[190] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 09/0 : 6[1d0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 11/0 : 4[1b0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 10/0 : 2[190] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 05/0 : 3[1a0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 00/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 11/0 : 3[1a0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 01/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 02/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 05/0 : 6[1d0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 06/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 00/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 05/0 : 1[180] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 03/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 11/0 : 6[1d0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 07/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 01/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 08/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 06/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 11/0 : 1[180] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 04/0 : 5[1c0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 09/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 07/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 10/0 : 5[1c0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 05/0 : 2[190] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 04/0 : 6[1d0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 02/0 : 4[1b0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 11/0 : 2[190] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 04/0 : 2[190] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 10/0 : 6[1d0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 03/0 : 4[1b0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 04/0 : 7[1e0] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 10/0 : 2[190] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 08/0 : 4[1b0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 05/0 : 3[1a0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 00/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 10/0 : 7[1e0] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 09/0 : 4[1b0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 11/0 : 3[1a0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 01/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 02/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 06/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 05/0 : 6[1d0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 03/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 00/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 11/0 : 6[1d0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 07/0 : 4[1b0] -> 7[1e0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 08/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 01/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 09/0 : 1[180] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 06/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 07/0 : 5[1c0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 02/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 00/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 03/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 01/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 02/0 : 4[1b0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 05/0 : 2[190] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 04/0 : 6[1d0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 08/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 06/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 03/0 : 4[1b0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 11/0 : 2[190] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 10/0 : 6[1d0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 07/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 09/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 04/0 : 7[1e0] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 08/0 : 4[1b0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 10/0 : 7[1e0] -> 3[1a0] via P2P/IPC\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 09/0 : 4[1b0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 04/0 : 3[1a0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 05/0 : 7[1e0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 10/0 : 3[1a0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 11/0 : 7[1e0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 02/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 00/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 02/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 03/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 01/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 08/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 06/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 03/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 07/0 : 3[1a0] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 09/0 : 7[1e0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 00/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 08/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 01/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 09/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 04/0 : 3[1a0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 05/0 : 7[1e0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 06/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 02/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 10/0 : 3[1a0] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 11/0 : 7[1e0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 04/0 : 1[180] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 07/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 03/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 05/0 : 5[1c0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 00/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 10/0 : 1[180] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 08/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 11/0 : 5[1c0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 01/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 09/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 06/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Connected all trees\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO threadThresholds 8/8/64 | 64/8/64 | 512 | 512\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 02/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO 12 coll channels, 16 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 00/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 07/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Connected all trees\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO threadThresholds 8/8/64 | 64/8/64 | 512 | 512\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO 12 coll channels, 16 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 03/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 01/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Connected all trees\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO threadThresholds 8/8/64 | 64/8/64 | 512 | 512\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO 12 coll channels, 16 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 08/1 : 3[1a0] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 06/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 08/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 07/0 : 7[1e0] -> 6[1d0] via P2P/IPC\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 09/0 : 3[1a0] -> 2[190] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 02/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 04/0 : 1[180] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 00/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 05/0 : 5[1c0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Connected all trees\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO threadThresholds 8/8/64 | 64/8/64 | 512 | 512\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 03/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO 12 coll channels, 16 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Connected all trees\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO threadThresholds 8/8/64 | 64/8/64 | 512 | 512\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO 12 coll channels, 16 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 08/1 : 7[1e0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 11/0 : 5[1c0] -> 4[1b0] via P2P/IPC\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 10/0 : 1[180] -> 0[170] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 01/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 08/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 06/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 09/0 : 2[190] -> 1[180] via P2P/IPC\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Connected all trees\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO threadThresholds 8/8/64 | 64/8/64 | 512 | 512\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO 12 coll channels, 16 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 04/1 : 2[190] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 07/0 : 6[1d0] -> 5[1c0] via P2P/IPC\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Connected all trees\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO threadThresholds 8/8/64 | 64/8/64 | 512 | 512\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO 12 coll channels, 16 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 09/1 : 3[1a0] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Connected all trees\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO threadThresholds 8/8/64 | 64/8/64 | 512 | 512\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO 12 coll channels, 16 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Connected all trees\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO threadThresholds 8/8/64 | 64/8/64 | 512 | 512\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO 12 coll channels, 16 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Connected all trees\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO threadThresholds 8/8/64 | 64/8/64 | 512 | 512\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO 12 coll channels, 16 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 04/1 : 6[1d0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 09/1 : 7[1e0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 05/1 : 2[190] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Connected all trees\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO threadThresholds 8/8/64 | 64/8/64 | 512 | 512\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO 12 coll channels, 16 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 08/1 : 3[1a0] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 04/1 : 3[1a0] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 04/1 : 7[1e0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Connected all trees\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO threadThresholds 8/8/64 | 64/8/64 | 512 | 512\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO 12 coll channels, 16 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 08/1 : 7[1e0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 05/1 : 6[1d0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Connected all trees\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO threadThresholds 8/8/64 | 64/8/64 | 512 | 512\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO 12 coll channels, 16 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Connected all trees\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO threadThresholds 8/8/64 | 64/8/64 | 512 | 512\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO 12 coll channels, 16 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 05/1 : 3[1a0] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 05/1 : 7[1e0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Connected all trees\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO threadThresholds 8/8/64 | 64/8/64 | 512 | 512\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO 12 coll channels, 16 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 04/1 : 2[190] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Connected all trees\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO threadThresholds 8/8/64 | 64/8/64 | 512 | 512\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO 12 coll channels, 16 p2p channels, 2 p2p channels per peer\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 04/1 : 6[1d0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 09/1 : 7[1e0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 09/1 : 3[1a0] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 12/1 : 3[1a0] -> 6[1d0] via P2P/indirect/7[1e0]\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 04/1 : 7[1e0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 12/1 : 7[1e0] -> 2[190] via P2P/indirect/3[1a0]\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 05/1 : 6[1d0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 05/1 : 2[190] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 04/1 : 3[1a0] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 05/1 : 7[1e0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO Channel 13/1 : 3[1a0] -> 6[1d0] via P2P/indirect/7[1e0]\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 12/1 : 5[1c0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO Channel 13/1 : 7[1e0] -> 2[190] via P2P/indirect/3[1a0]\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 12/1 : 6[1d0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 12/1 : 7[1e0] -> 2[190] via P2P/indirect/3[1a0]\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 05/1 : 3[1a0] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 12/1 : 1[180] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 13/1 : 5[1c0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 12/1 : 2[190] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 13/1 : 6[1d0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 12/1 : 3[1a0] -> 6[1d0] via P2P/indirect/7[1e0]\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 10/1 : 0[170] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO Channel 13/1 : 7[1e0] -> 2[190] via P2P/indirect/3[1a0]\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 13/1 : 1[180] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 13/1 : 2[190] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 12/1 : 6[1d0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 11/1 : 0[170] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO Channel 13/1 : 3[1a0] -> 6[1d0] via P2P/indirect/7[1e0]\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 10/1 : 4[1b0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 12/1 : 1[180] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 12/1 : 2[190] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 12/1 : 5[1c0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 11/1 : 4[1b0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 13/1 : 6[1d0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 13/1 : 1[180] -> 4[1b0] via P2P/indirect/0[170]\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 13/1 : 5[1c0] -> 0[170] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 13/1 : 2[190] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 10/1 : 4[1b0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 10/1 : 1[180] -> 6[1d0] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 10/1 : 6[1d0] -> 3[1a0] via P2P/indirect/2[190]\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 10/1 : 2[190] -> 7[1e0] via P2P/indirect/6[1d0]\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 11/1 : 4[1b0] -> 1[180] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 11/1 : 1[180] -> 6[1d0] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 10/1 : 0[170] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 10/1 : 5[1c0] -> 2[190] via P2P/indirect/1[180]\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO Channel 11/1 : 6[1d0] -> 3[1a0] via P2P/indirect/2[190]\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO Channel 11/1 : 2[190] -> 7[1e0] via P2P/indirect/6[1d0]\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 11/1 : 5[1c0] -> 2[190] via P2P/indirect/1[180]\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 06/1 : 4[1b0] -> 2[190] via P2P/indirect/6[1d0]\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 11/1 : 0[170] -> 5[1c0] via P2P/indirect/1[180]\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 10/1 : 6[1d0] -> 3[1a0] via P2P/indirect/2[190]\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 10/1 : 1[180] -> 6[1d0] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 06/1 : 0[170] -> 6[1d0] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 11/1 : 1[180] -> 6[1d0] via P2P/indirect/5[1c0]\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO Channel 11/1 : 6[1d0] -> 3[1a0] via P2P/indirect/2[190]\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 06/1 : 1[180] -> 7[1e0] via P2P/indirect/3[1a0]\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 07/1 : 4[1b0] -> 2[190] via P2P/indirect/6[1d0]\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 10/1 : 2[190] -> 7[1e0] via P2P/indirect/6[1d0]\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 07/1 : 0[170] -> 6[1d0] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 06/1 : 4[1b0] -> 2[190] via P2P/indirect/6[1d0]\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 10/1 : 5[1c0] -> 2[190] via P2P/indirect/1[180]\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 06/1 : 5[1c0] -> 3[1a0] via P2P/indirect/1[180]\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO Channel 11/1 : 2[190] -> 7[1e0] via P2P/indirect/6[1d0]\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO Channel 07/1 : 1[180] -> 7[1e0] via P2P/indirect/3[1a0]\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 06/1 : 1[180] -> 7[1e0] via P2P/indirect/3[1a0]\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 07/1 : 4[1b0] -> 2[190] via P2P/indirect/6[1d0]\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 11/1 : 5[1c0] -> 2[190] via P2P/indirect/1[180]\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO Channel 07/1 : 5[1c0] -> 3[1a0] via P2P/indirect/1[180]\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO Channel 07/1 : 1[180] -> 7[1e0] via P2P/indirect/3[1a0]\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 14/1 : 0[170] -> 7[1e0] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 14/1 : 4[1b0] -> 3[1a0] via P2P/indirect/0[170]\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO Channel 15/1 : 0[170] -> 7[1e0] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 06/1 : 0[170] -> 6[1d0] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO Channel 15/1 : 4[1b0] -> 3[1a0] via P2P/indirect/0[170]\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 06/1 : 5[1c0] -> 3[1a0] via P2P/indirect/1[180]\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 07/1 : 0[170] -> 6[1d0] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO Channel 07/1 : 5[1c0] -> 3[1a0] via P2P/indirect/1[180]\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 14/1 : 0[170] -> 7[1e0] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 14/1 : 4[1b0] -> 3[1a0] via P2P/indirect/0[170]\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO Channel 15/1 : 0[170] -> 7[1e0] via P2P/indirect/4[1b0]\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO Channel 15/1 : 4[1b0] -> 3[1a0] via P2P/indirect/0[170]\n",
      "[1,mpirank:7,algo-1]<stdout>:algo-1:205:205 [7] NCCL INFO comm 0x5626de8d0c90 rank 7 nranks 8 cudaDev 7 busId 1e0 - Init COMPLETE\n",
      "[1,mpirank:3,algo-1]<stdout>:algo-1:211:211 [3] NCCL INFO comm 0x555de59d6740 rank 3 nranks 8 cudaDev 3 busId 1a0 - Init COMPLETE\n",
      "[1,mpirank:4,algo-1]<stdout>:algo-1:207:207 [4] NCCL INFO comm 0x55bb939f9690 rank 4 nranks 8 cudaDev 4 busId 1b0 - Init COMPLETE\n",
      "[1,mpirank:6,algo-1]<stdout>:algo-1:204:204 [6] NCCL INFO comm 0x55a151d32520 rank 6 nranks 8 cudaDev 6 busId 1d0 - Init COMPLETE\n",
      "[1,mpirank:2,algo-1]<stdout>:algo-1:201:201 [2] NCCL INFO comm 0x55a8dd832260 rank 2 nranks 8 cudaDev 2 busId 190 - Init COMPLETE\n",
      "[1,mpirank:0,algo-1]<stdout>:algo-1:223:223 [0] NCCL INFO comm 0x55850f6b5620 rank 0 nranks 8 cudaDev 0 busId 170 - Init COMPLETE\n",
      "[1,mpirank:0,algo-1]<stdout>:Running smdistributed.dataparallel v1.7.0\n",
      "[1,mpirank:0,algo-1]<stdout>:SMDDP: Multi node ENA mode\n",
      "[1,mpirank:1,algo-1]<stdout>:algo-1:213:213 [1] NCCL INFO comm 0x55a2b7fbfc30 rank 1 nranks 8 cudaDev 1 busId 180 - Init COMPLETE\n",
      "[1,mpirank:5,algo-1]<stdout>:algo-1:209:209 [5] NCCL INFO comm 0x5588bebdadd0 rank 5 nranks 8 cudaDev 5 busId 1c0 - Init COMPLETE\n",
      "[1,mpirank:15,algo-2]<stdout>:algo-2:139:139 [7] NCCL INFO comm 0x559631fdcb70 rank 7 nranks 8 cudaDev 7 busId 1e0 - Init COMPLETE\n",
      "[1,mpirank:12,algo-2]<stdout>:algo-2:145:145 [4] NCCL INFO comm 0x55a11029e610 rank 4 nranks 8 cudaDev 4 busId 1b0 - Init COMPLETE\n",
      "[1,mpirank:8,algo-2]<stdout>:algo-2:159:159 [0] NCCL INFO comm 0x55c199b8fb00 rank 0 nranks 8 cudaDev 0 busId 170 - Init COMPLETE\n",
      "[1,mpirank:13,algo-2]<stdout>:algo-2:149:149 [5] NCCL INFO comm 0x55ec60d64680 rank 5 nranks 8 cudaDev 5 busId 1c0 - Init COMPLETE\n",
      "[1,mpirank:10,algo-2]<stdout>:algo-2:148:148 [2] NCCL INFO comm 0x55e6c5875610 rank 2 nranks 8 cudaDev 2 busId 190 - Init COMPLETE\n",
      "[1,mpirank:11,algo-2]<stdout>:algo-2:143:143 [3] NCCL INFO comm 0x55e18b1773f0 rank 3 nranks 8 cudaDev 3 busId 1a0 - Init COMPLETE\n",
      "[1,mpirank:9,algo-2]<stdout>:algo-2:137:137 [1] NCCL INFO comm 0x55f352c57080 rank 1 nranks 8 cudaDev 1 busId 180 - Init COMPLETE\n",
      "[1,mpirank:14,algo-2]<stdout>:algo-2:141:141 [6] NCCL INFO comm 0x55e37afe6c60 rank 6 nranks 8 cudaDev 6 busId 1d0 - Init COMPLETE\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,     1] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,    21] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,    41] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,    61] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,    81] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   101] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   121] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   141] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   161] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   181] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   201] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   221] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   241] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   261] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   281] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   301] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   321] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   341] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   361] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   381] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   401] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   421] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   441] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   461] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   481] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   501] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   521] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   541] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   561] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   581] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   601] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   621] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   641] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   661] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   681] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   701] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   721] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   741] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   761] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   781] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   801] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   821] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   841] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   861] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   881] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   901] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   921] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   941] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   961] loss: 0.001\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,   981] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,  1001] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,  1021] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,  1041] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,  1061] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,  1081] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,  1101] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,  1121] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,  1141] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,  1161] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,  1181] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,  1201] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,  1221] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 1,  1241] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,     1] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,    21] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,    41] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,    61] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,    81] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   101] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   121] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   141] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   161] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   181] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   201] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   221] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   241] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   261] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   281] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   301] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   321] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   341] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   361] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   381] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   401] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   421] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   441] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   461] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   481] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   501] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   521] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   541] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   561] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   581] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   601] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   621] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   641] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   661] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   681] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   701] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   721] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   741] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   761] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   781] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   801] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   821] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   841] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   861] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   881] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   901] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   921] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   941] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   961] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,   981] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,  1001] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,  1021] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,  1041] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,  1061] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,  1081] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,  1101] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,  1121] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,  1141] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,  1161] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,  1181] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,  1201] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,  1221] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:[epoch: 2,  1241] loss: 0.000\n",
      "[1,mpirank:0,algo-1]<stdout>:Finished Training\n",
      "2023-02-02 20:50:24,728 sagemaker-training-toolkit INFO     Invoked on_terminate from psutil.wait_for_procs\n",
      "2023-02-02 20:50:24,728 sagemaker-training-toolkit INFO     process psutil.Process(pid=63, name='orted', status='terminated', started='20:49:26') terminated with exit code None\n",
      "2023-02-02 20:50:24,728 sagemaker-training-toolkit INFO     Reporting status for ORTEd process. gone: [psutil.Process(pid=63, name='orted', status='terminated', started='20:49:26')] alive: []\n",
      "2023-02-02 20:50:24,728 sagemaker-training-toolkit INFO     Orted process exited\n",
      "2023-02-02 20:50:24,703 sagemaker-training-toolkit INFO     Waiting for the process to finish and give a return code.\n",
      "2023-02-02 20:50:24,704 sagemaker-training-toolkit INFO     Done waiting for a return code. Received 0 from exiting process.\n",
      "2023-02-02 20:50:24,704 sagemaker-training-toolkit INFO     Begin writing status file from leader node to worker nodes\n",
      "2023-02-02 20:50:24,704 sagemaker-training-toolkit INFO     Start writing mpirun finished status to algo-2\n",
      "2023-02-02 20:50:24,864 sagemaker-training-toolkit INFO     output from subprocess run CompletedProcess(args=['ssh', 'algo-2', 'touch', '/tmp/done.algo-1'], returncode=0, stdout='', stderr='')\n",
      "2023-02-02 20:50:24,864 sagemaker-training-toolkit INFO     Finished writing status file\n",
      "2023-02-02 20:50:54,759 sagemaker-training-toolkit INFO     Begin looking for status file on algo-2\n",
      "2023-02-02 20:50:54,759 sagemaker-training-toolkit INFO     MPI training job status file found. Exit gracefully\n",
      "2023-02-02 20:50:54,759 sagemaker-training-toolkit INFO     End looking for status file\n",
      "2023-02-02 20:50:54,759 sagemaker-training-toolkit INFO     MPI process finished.\n",
      "2023-02-02 20:50:54,759 sagemaker-training-toolkit INFO     Reporting training SUCCESS\n",
      "2023-02-02 20:50:54,894 sagemaker-training-toolkit INFO     Finished writing status file from leader node to worker nodes\n",
      "2023-02-02 20:50:54,894 sagemaker-training-toolkit INFO     Reporting training SUCCESS\n",
      "\n",
      "2023-02-02 20:51:12 Uploading - Uploading generated training model\n",
      "2023-02-02 20:51:12 Completed - Training job completed\n",
      "Training seconds: 756\n",
      "Billable seconds: 756\n"
     ]
    }
   ],
   "source": [
    "import sagemaker\n",
    "from sagemaker.pytorch import PyTorch\n",
    "\n",
    "session = sagemaker.Session()\n",
    "role = sagemaker.get_execution_role(sagemaker_session=session)\n",
    "\n",
    "estimator = PyTorch(\n",
    "    role=role,\n",
    "    source_dir=\"sagemaker\",\n",
    "    entry_point=\"sagemaker_distributed_reading_service.py\",\n",
    "    image_uri=\"763104351884.dkr.ecr.us-west-2.amazonaws.com/pytorch-training:1.13.1-gpu-py39-cu117-ubuntu20.04-sagemaker\",\n",
    "    instance_count=2,\n",
    "    # Instance types supported by the SageMaker data parallel library: \n",
    "    # ml.p4d.24xlarge, ml.p3dn.24xlarge, and ml.p3.16xlarge\n",
    "    instance_type=\"ml.p3.16xlarge\",\n",
    "    # Training using the SageMaker data parallel distributed training strategy\n",
    "    distribution={ \"smdistributed\": { \"dataparallel\": { \"enabled\": True } } }\n",
    ")\n",
    "\n",
    "estimator.fit()"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## S3FileLoader <a class=\"anchor\" id=\"35\"></a>\n",
    "\n",
    "TorchData now provides a new [`S3FileLoader`](https://pytorch.org/data/main/generated/torchdata.datapipes.iter.S3FileLoader.html) Iterable DataPipe that loads Amazon S3 files from the given S3 URLs.\n",
    "\n",
    "Note that the below example of `S3FileLoader` isn't necessary if using Amazon SageMaker's Fast File Mode. Fast File Mode exposes S3 objects using a POSIX-compliant file system interface. This means that data can be used as if it were available on your local system. It streams this data on demand, and performs better when files are read sequentially."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:botocore.credentials:Found credentials from IAM Role: BaseNotebookInstanceEc2InstanceRole\n"
     ]
    }
   ],
   "source": [
    "from sagemaker import Session\n",
    "\n",
    "session = Session()\n",
    "bucket = session.default_bucket()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "upload: data/vision/webdataset/idx2label to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/idx2label\n",
      "upload: data/vision/webdataset/train-000006.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000006.tar\n",
      "upload: data/vision/webdataset/train-000003.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000003.tar\n",
      "upload: data/vision/webdataset/train-000008.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000008.tar\n",
      "upload: data/vision/webdataset/train-000004.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000004.tar\n",
      "upload: data/vision/webdataset/train-000007.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000007.tar\n",
      "upload: data/vision/webdataset/train-000002.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000002.tar\n",
      "upload: data/vision/webdataset/train-000005.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000005.tar\n",
      "upload: data/vision/webdataset/train-000001.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000001.tar\n",
      "upload: data/vision/webdataset/train-000009.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000009.tar\n",
      "upload: data/vision/webdataset/train-000010.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000010.tar\n",
      "upload: data/vision/webdataset/train-000000.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000000.tar\n",
      "upload: data/vision/webdataset/train-000012.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000012.tar\n",
      "upload: data/vision/webdataset/train-000013.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000013.tar\n",
      "upload: data/vision/webdataset/train-000014.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000014.tar\n",
      "upload: data/vision/webdataset/train-000011.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000011.tar\n",
      "upload: data/vision/webdataset/train-000015.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000015.tar\n",
      "upload: data/vision/webdataset/train-000017.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000017.tar\n",
      "upload: data/vision/webdataset/train-000016.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000016.tar\n",
      "upload: data/vision/webdataset/train-000019.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000019.tar\n",
      "upload: data/vision/webdataset/train-000018.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000018.tar\n",
      "upload: data/vision/webdataset/train-000020.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000020.tar\n",
      "upload: data/vision/webdataset/train-000022.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000022.tar\n",
      "upload: data/vision/webdataset/train-000023.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000023.tar\n",
      "upload: data/vision/webdataset/train-000021.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000021.tar\n",
      "upload: data/vision/webdataset/train-000025.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000025.tar\n",
      "upload: data/vision/webdataset/train-000024.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000024.tar\n",
      "upload: data/vision/webdataset/train-000026.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000026.tar\n",
      "upload: data/vision/webdataset/train-000027.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000027.tar\n",
      "upload: data/vision/webdataset/train-000028.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000028.tar\n",
      "upload: data/vision/webdataset/train-000029.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000029.tar\n",
      "upload: data/vision/webdataset/train-000031.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000031.tar\n",
      "upload: data/vision/webdataset/train-000030.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000030.tar\n",
      "upload: data/vision/webdataset/train-000032.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000032.tar\n",
      "upload: data/vision/webdataset/train-000033.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000033.tar\n",
      "upload: data/vision/webdataset/train-000038.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000038.tar\n",
      "upload: data/vision/webdataset/train-000035.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000035.tar\n",
      "upload: data/vision/webdataset/train-000034.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000034.tar\n",
      "upload: data/vision/webdataset/train-000039.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000039.tar\n",
      "upload: data/vision/webdataset/train-000036.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000036.tar\n",
      "upload: data/vision/webdataset/train-000037.tar to s3://sagemaker-us-west-2-197237853439/data/vision/webdataset/train-000037.tar\n"
     ]
    }
   ],
   "source": [
    "# Upload data to S3\n",
    "!aws s3 sync data/vision/webdataset s3://$bucket/data/vision/webdataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<PIL.Image.Image image mode=RGB size=256x256 at 0x7FF5B115DA30>"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from functools import partial\n",
    "\n",
    "import torch\n",
    "import torchvision.transforms as transforms\n",
    "from torchdata.datapipes.iter import IterableWrapper, S3FileLoader\n",
    "from webdataset.autodecode import imagehandler\n",
    "\n",
    "\n",
    "device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
    "\n",
    "# Define image decoders\n",
    "pil_handler = imagehandler('pil')\n",
    "\n",
    "# Define simple transforms (since resizing is required to collate)\n",
    "data_transform = transforms.Compose([transforms.RandomResizedCrop(256),])\n",
    "\n",
    "# Define how to decode webdataset keys\n",
    "def decode(item, transform=data_transform, image_decoder=pil_handler):\n",
    "    key, value = item\n",
    "    if key.endswith(\".cls\"):\n",
    "        class_id = int(value.read().decode(\"utf-8\"))\n",
    "        return key, class_id\n",
    "    elif key.endswith(\".jpg\"):\n",
    "        image = image_decoder(\".jpg\", value.read())\n",
    "        return key, transform(image)\n",
    "\n",
    "# Define simple collate function\n",
    "def collate_fn(batch, device=device):\n",
    "    classes = torch.Tensor([sample['.cls'] for sample in batch]).to(device)\n",
    "    images = torch.stack([sample['.jpg'] for sample in batch]).to(device)\n",
    "    return classes, images\n",
    "\n",
    "# Create pipeline\n",
    "dp_s3_urls = IterableWrapper([f's3://{bucket}/data/vision/webdataset',]).list_files_by_s3()\n",
    "dp = (dp_s3_urls.filter(lambda x: x.endswith(\".tar\"))\n",
    "      .shuffle()\n",
    "      .sharding_filter()\n",
    "      .load_files_by_s3()\n",
    "      .load_from_tar()\n",
    "      .map(decode)\n",
    "      .webdataset()\n",
    "      #.batch(8)\n",
    "      #.collate()\n",
    ")\n",
    "\n",
    "next(iter(dp))['.jpg']"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----\n",
    "\n",
    "## AIStore on S3 [[13]](https://github.com/NVIDIA/aistore/blob/master/python/aistore/examples/aisio-pytorch/aisio_pytorch_example.ipynb) <a class=\"anchor\" id=\"36\"></a>\n",
    "\n",
    "[AIStore](https://github.com/NVIDIA/aistore#features) is a lightweight object storage system with the capability to linearly scale out with each added storage node and a special focus on petascale deep learning.\n",
    "\n",
    "Paper: [High Performance I/O for Large Scale Deep Learning](https://arxiv.org/abs/2001.01858)\n",
    "\n",
    "Blog [2020]: [Efficient PyTorch I/O library for Large Datasets, Many Files, Many GPUs](https://pytorch.org/blog/efficient-pytorch-io-library-for-large-datasets-many-files-many-gpus)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Running the AIStore Cluster\n",
    "\n",
    "[AIStore](https://github.com/NVIDIA/aistore) (AIS for short) is a highly available lightweight object storage system that specifically focuses on petascale deep learning. As a reliable redundant storage, AIS supports n-way mirroring and erasure coding. But it is not purely – or not only – a storage system: it’ll shuffle user datasets and run custom extract-transform-load workloads.\n",
    "\n",
    "AIS is an elastic cluster that can grow and shrink at runtime and can be ad-hoc deployed, with or without Kubernetes, anywhere from a single Linux machine to a bare-metal cluster of any size. AIS fully supports Amazon S3, Google Cloud, and Microsoft Azure backends, providing a unified namespace across multiple connected backends and/or other AIS clusters, and [more](https://github.com/NVIDIA/aistore#features).\n",
    "\n",
    "[Getting started with AIS](https://github.com/NVIDIA/aistore/blob/master/docs/getting_started.md) will take only a few minutes (prerequisites boil down to having a Linux with a disk) and can be done either by running a prebuilt [all-in-one docker image or directly from the open-source](https://github.com/NVIDIA/aistore/tree/master/deploy).\n",
    "\n",
    "To keep this example simple, we will be running a [minimal standalone docker deployment of AIStore](https://github.com/NVIDIA/aistore/blob/master/deploy/prod/docker/single/README.md)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3a7920f869d890f17ce1bc624c9df56182bf016078af7f58be349e66e1a20fe3\n"
     ]
    }
   ],
   "source": [
    "# Running the AIStore cluster in a container on port 51080\n",
    "# Note: The mounted path should have enough space to load the dataset\n",
    "\n",
    "! docker run -d \\\n",
    "    -p 51080:51080 \\\n",
    "    -e AWS_ACCESS_KEY_ID=\"FILL IN\" \\\n",
    "    -e AWS_SECRET_ACCESS_KEY=\"FILL IN\" \\\n",
    "    -e AWS_DEFAULT_REGION=\"us-west-2\" \\\n",
    "    -e AIS_BACKEND_PROVIDERS=\"aws\" \\\n",
    "    --name aistore \\\n",
    "    -v /disk0:/ais/disk0 \\\n",
    "    aistore/cluster-minimal:latest"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To create and put objects (dataset) in the bucket, we can use the [AIS CLI](https://github.com/NVIDIA/aistore/blob/master/docs/cli.md), the [Python SDK](https://github.com/NVIDIA/aistore/tree/master/python/aistore), or a combination of both."
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In order to interact with your running AIS instance, you will need to create a client object:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "from aistore import Client\n",
    "\n",
    "client = Client(\"http://localhost:51080\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 54,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Check if AIS is deployed and running\n",
    "client.cluster().is_aistore_running()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Smap(tmap={'TnoCEbzp': Snode(daemon_id='TnoCEbzp', daemon_type='target', public_net=NetInfo(node_hostname='', daemon_port='51081', direct_url='http://172.17.0.2:51081'), intra_control_net=NetInfo(node_hostname='', daemon_port='51081', direct_url='http://172.17.0.2:51081'), intra_data_net=NetInfo(node_hostname='', daemon_port='51081', direct_url='http://172.17.0.2:51081'), flags=0)}, pmap={'BunPIMbV': Snode(daemon_id='BunPIMbV', daemon_type='proxy', public_net=NetInfo(node_hostname='', daemon_port='51080', direct_url='http://172.17.0.2:51080'), intra_control_net=NetInfo(node_hostname='', daemon_port='51080', direct_url='http://172.17.0.2:51080'), intra_data_net=NetInfo(node_hostname='', daemon_port='51080', direct_url='http://172.17.0.2:51080'), flags=2)}, proxy_si=Snode(daemon_id='BunPIMbV', daemon_type='proxy', public_net=NetInfo(node_hostname='', daemon_port='51080', direct_url='http://172.17.0.2:51080'), intra_control_net=NetInfo(node_hostname='', daemon_port='51080', direct_url='http://172.17.0.2:51080'), intra_data_net=NetInfo(node_hostname='', daemon_port='51080', direct_url='http://172.17.0.2:51080'), flags=2), version=4, uuid='Z_w_gBBKdn', creation_time='2023-02-01 22:42:31.999876972 +0000 UTC m=+19.031811442')"
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Get cluster information\n",
    "client.cluster().get_info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create a bucket named \"cifar10\"\n",
    "bucket = \"cifar10\"\n",
    "client.bucket(bucket).create()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz to /home/ec2-user/SageMaker/torchdata-examples/data/vision/cifar10/cifar-10-python.tar.gz\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 170498071/170498071 [00:05<00:00, 31665734.55it/s]\n"
     ]
    }
   ],
   "source": [
    "# Download the dataset to the local machine\n",
    "from torchdata.datapipes.iter import IterDataPipe\n",
    "from torchvision.prototype import datasets\n",
    "\n",
    "root = \"data/vision/cifar10\"\n",
    "split = \"train\"\n",
    "train = datasets.load(\"cifar10\", root=root, split=split)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Files to upload:\n",
      "EXTENSION\t COUNT\t SIZE\n",
      ".gz\t\t 1\t 162.60MiB\n",
      "TOTAL\t\t1\t162.60MiB\n",
      "PUT 1 object to \"ais://cifar10\"\n"
     ]
    }
   ],
   "source": [
    "# Put in aistore\n",
    "! ais config cli set cluster.url=http://localhost:51080\n",
    "! ais object put -r -y data/vision/cifar10 ais://cifar10/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "BucketList(uuid='J1eS_BFKon', entries=[BucketEntry(name='cifar-10-python.tar.gz', size=170498071, checksum='68c32b7196075cb8', atime='01 Feb 23 22:49 UTC', version='', target_url='', copies=0, flags=64)], continuation_token='', flags=0)"
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "client.bucket(bucket).list_objects()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### PyTorch Integration\n",
    "\n",
    "You can list and load data from AIS buckets (buckets that are not 3rd party backend-based) and remote cloud buckets (3rd party backend-based cloud buckets) in PyTorch using `AISFileLister` and `AISFileLoader`.\n",
    "\n",
    "`AISFileLister` and `AISFileLoader` are now available as a part of official pytorch/data project.\n",
    "\n",
    "\n",
    "```python\n",
    "    from torchdata.datapipes.iter import AISFileLister, AISFileLoader\n",
    "\n",
    "    # provide list of prefixes to load and list data from\n",
    "    ais_prefixes = ['gcp://bucket-name/folder/', 'aws:bucket-name/folder/', 'ais://bucket-name/folder/', ...]\n",
    "\n",
    "    # List all files for these prefixes using AISFileLister\n",
    "    dp_ais_urls = AISFileLister(url='localhost:8080', source_datapipe=ais_prefixes)\n",
    "\n",
    "    # print(list(dp_ais_urls))\n",
    "\n",
    "    # Load files using AISFileLoader\n",
    "    dp_files = AISFileLoader(url='localhost:8080', source_datapipe=dp_ais_urls)\n",
    "\n",
    "    for url, file in dp_files:\n",
    "        pass\n",
    "```"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Preloaded dataset\n",
    "\n",
    "The following assumes that AIS cluster is running and one of its buckets contains Caltech-256 dataset."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['ais://cifar10/cifar-10-python.tar.gz']\n"
     ]
    }
   ],
   "source": [
    "from torchdata.datapipes.iter import AISFileLister, AISFileLoader\n",
    "\n",
    "# list of prefixes which contain data\n",
    "image_prefix = [\"ais://cifar10/\"]\n",
    "\n",
    "# Listing all files starting with these prefixes on AIStore\n",
    "dp_urls = AISFileLister(url=\"http://localhost:51080\", source_datapipe=image_prefix)\n",
    "\n",
    "# list obj urls\n",
    "print(list(dp_urls))\n",
    "\n",
    "# loading data using AISFileLoader\n",
    "dp = AISFileLoader(url=\"http://localhost:51080\", source_datapipe=dp_urls)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Since cifar10 is small, we only see one file above. \n",
    "\n",
    "Next we'll build a DataPipe to load torchvision's cifar10 `tar.gz` data. Check this out in more detail [here](https://github.com/pytorch/vision/blob/main/torchvision/prototype/datasets/_builtin/cifar.py)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import io\n",
    "import pathlib\n",
    "import pickle\n",
    "from typing import Any, BinaryIO, cast, Dict, Iterator, List, Optional, Tuple, Union\n",
    "\n",
    "import numpy as np\n",
    "import torch\n",
    "from torchdata.datapipes.iter import TarArchiveLoader, Filter, IterDataPipe, Mapper\n",
    "from torchvision.prototype.datapoints import Image, Label\n",
    "from torchvision.prototype.datasets.utils._internal import (\n",
    "    hint_sharding,\n",
    "    hint_shuffling,\n",
    "    path_comparator,\n",
    "    read_categories_file,\n",
    ")\n",
    "\n",
    "class CifarFileReader(IterDataPipe[Tuple[np.ndarray, int]]):\n",
    "    def __init__(self, datapipe: IterDataPipe[Dict[str, Any]], *, labels_key: str) -> None:\n",
    "        self.datapipe = datapipe\n",
    "        self.labels_key = labels_key\n",
    "\n",
    "    def __iter__(self) -> Iterator[Tuple[np.ndarray, int]]:\n",
    "        for mapping in self.datapipe:\n",
    "            image_arrays = mapping[\"data\"].reshape((-1, 3, 32, 32))\n",
    "            category_idcs = mapping[self.labels_key]\n",
    "            yield from iter(zip(image_arrays, category_idcs))\n",
    "\n",
    "categories = ['airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck']\n",
    "\n",
    "split = \"train\"\n",
    "\n",
    "def _is_data_file(data: Tuple[str, Any]) -> bool:\n",
    "    path = pathlib.Path(data[0])\n",
    "    return path.name.startswith(\"data\" if split == \"train\" else \"test\")\n",
    "\n",
    "def _unpickle(data: Tuple[str, io.BytesIO]) -> Dict[str, Any]:\n",
    "    _, file = data\n",
    "    content = cast(Dict[str, Any], pickle.load(file, encoding=\"latin1\"))\n",
    "    file.close()\n",
    "    return content\n",
    "\n",
    "def _prepare_sample(data: Tuple[np.ndarray, int]) -> Dict[str, Any]:\n",
    "    image_array, category_idx = data\n",
    "    return dict(\n",
    "        image=Image(image_array),\n",
    "        label=Label(category_idx, categories=categories),\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dict_keys(['image', 'label'])\n",
      "torch.Size([8, 3, 32, 32])\n"
     ]
    }
   ],
   "source": [
    "device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
    "\n",
    "dp = TarArchiveLoader(dp)\n",
    "dp = Filter(dp, _is_data_file)\n",
    "dp = Mapper(dp, _unpickle)\n",
    "dp = CifarFileReader(dp, labels_key=\"labels\")\n",
    "dp = hint_shuffling(dp)\n",
    "dp = hint_sharding(dp)\n",
    "dp = Mapper(dp, _prepare_sample)\n",
    "dp = dp.batch(8).collate()\n",
    "\n",
    "for x in dp:\n",
    "    print(x.keys())\n",
    "    images = x['image'].to(device)\n",
    "    labels = x['label'].to(device)\n",
    "    print(images.shape)\n",
    "    break"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<PIL.Image.Image image mode=RGB size=32x32 at 0x7FEF5EFC62E0>"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import torchvision.transforms as T\n",
    "\n",
    "T.ToPILImage()(images[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "client.bucket(bucket).delete()"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Boto3 Compatibility\n",
    "\n",
    "To use boto3 or botocore as client libraries for AIStore:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "import boto3\n",
    "from aistore.botocore_patch import botocore"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----\n",
    "\n",
    "## References <a class=\"anchor\" id=\"37\"></a>\n",
    "\n",
    "- [1] https://github.com/pytorch/data/blob/main/README.md#torchdata\n",
    "- [2] https://github.com/pytorch/data/blob/main/README.md#why-composable-data-loading\n",
    "- [3] https://github.com/pytorch/data/blob/main/README.md#what-are-datapipes\n",
    "- [4] https://pytorch.org/data/main/dp_tutorial.html\n",
    "- [5] https://pytorch.org/data/0.5/tutorial.html\n",
    "- [6] https://github.com/pytorch/data#frequently-asked-questions-faq\n",
    "- [7] https://pytorch.org/data/beta/dataloader2.html#module-torchdata.dataloader2\n",
    "- [8] https://github.com/pytorch/vision/tree/main/torchvision/prototype/datasets/_builtin#how-do-i-compute-the-number-of-samples\n",
    "- [9] https://github.com/pytorch/torcharrow#torcharrow-a-data-processing-library-for-pytorch\n",
    "- [10] https://github.com/pytorch/data/blob/main/examples/vision/imagefolder.py\n",
    "- [11] https://github.com/pytorch/data/blob/main/examples/dataloader2/train_loop_torchtext.py\n",
    "- [12] https://pytorch.org/data/beta/tutorial.html#implementing-a-custom-datapipe\n",
    "- [13] https://github.com/NVIDIA/aistore/blob/master/python/aistore/examples/aisio-pytorch/aisio_pytorch_example.ipynb"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Thanks\n",
    "\n",
    "Special thanks to [Philip Meier](https://github.com/pmeier) and [Erjia Guan](https://github.com/ejguan) for answering several questions I posted in GitHub threads."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "torchdata",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.0"
  },
  "orig_nbformat": 4,
  "vscode": {
   "interpreter": {
    "hash": "ab0f1a42298be773a835694678f6f14e8b80491d0e5fdeae0e2faeed5ee0e17b"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}