[Bug]: The output of Aria model is not correct

[xffxff]

Your current environment

The output of `python collect_env.py`

Collecting environment information...
PyTorch version: 2.5.1+cu124
Is debug build: False
CUDA used to build PyTorch: 12.4
ROCM used to build PyTorch: N/A

OS: Ubuntu 22.04.4 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: Could not collect
CMake version: version 3.28.3
Libc version: glibc-2.35

Python version: 3.10.15 (main, Oct  3 2024, 07:27:34) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-5.15.0-91-generic-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: 12.4.99
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: 
GPU 0: NVIDIA H100 80GB HBM3
GPU 1: NVIDIA H100 80GB HBM3
GPU 2: NVIDIA H100 80GB HBM3
GPU 3: NVIDIA H100 80GB HBM3
GPU 4: NVIDIA H100 80GB HBM3
GPU 5: NVIDIA H100 80GB HBM3
GPU 6: NVIDIA H100 80GB HBM3
GPU 7: NVIDIA H100 80GB HBM3

Nvidia driver version: 535.161.08
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.9.0.0
/usr/lib/x86_64-linux-gnu/libcudnn_adv.so.9.0.0
/usr/lib/x86_64-linux-gnu/libcudnn_cnn.so.9.0.0
/usr/lib/x86_64-linux-gnu/libcudnn_engines_precompiled.so.9.0.0
/usr/lib/x86_64-linux-gnu/libcudnn_engines_runtime_compiled.so.9.0.0
/usr/lib/x86_64-linux-gnu/libcudnn_graph.so.9.0.0
/usr/lib/x86_64-linux-gnu/libcudnn_heuristic.so.9.0.0
/usr/lib/x86_64-linux-gnu/libcudnn_ops.so.9.0.0
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Architecture:                       x86_64
CPU op-mode(s):                     32-bit, 64-bit
Address sizes:                      46 bits physical, 57 bits virtual
Byte Order:                         Little Endian
CPU(s):                             192
On-line CPU(s) list:                0-191
Vendor ID:                          GenuineIntel
Model name:                         Intel(R) Xeon(R) Platinum 8468
CPU family:                         6
Model:                              143
Thread(s) per core:                 2
Core(s) per socket:                 48
Socket(s):                          2
Stepping:                           8
BogoMIPS:                           4200.00
Flags:                              fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq dtes64 monitor ds_cpl smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cat_l2 cdp_l3 invpcid_single intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local split_lock_detect avx_vnni avx512_bf16 wbnoinvd dtherm arat pln pts avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pconfig arch_lbr amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
L1d cache:                          4.5 MiB (96 instances)
L1i cache:                          3 MiB (96 instances)
L2 cache:                           192 MiB (96 instances)
L3 cache:                           210 MiB (2 instances)
NUMA node(s):                       2
NUMA node0 CPU(s):                  0-47,96-143
NUMA node1 CPU(s):                  48-95,144-191
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit:        Not affected
Vulnerability L1tf:                 Not affected
Vulnerability Mds:                  Not affected
Vulnerability Meltdown:             Not affected
Vulnerability Mmio stale data:      Not affected
Vulnerability Retbleed:             Not affected
Vulnerability Spec rstack overflow: Not affected
Vulnerability Spec store bypass:    Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1:           Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:           Mitigation; Enhanced IBRS, IBPB conditional, RSB filling, PBRSB-eIBRS SW sequence
Vulnerability Srbds:                Not affected
Vulnerability Tsx async abort:      Not affected

Versions of relevant libraries:
[pip3] flake8==7.1.1
[pip3] flashinfer==0.1.6+cu121torch2.4
[pip3] mypy==1.11.1
[pip3] mypy-extensions==1.0.0
[pip3] numpy==1.26.4
[pip3] nvidia-cublas-cu12==12.4.5.8
[pip3] nvidia-cuda-cupti-cu12==12.4.127
[pip3] nvidia-cuda-nvrtc-cu12==12.4.127
[pip3] nvidia-cuda-runtime-cu12==12.4.127
[pip3] nvidia-cudnn-cu12==9.1.0.70
[pip3] nvidia-cufft-cu12==11.2.1.3
[pip3] nvidia-curand-cu12==10.3.5.147
[pip3] nvidia-cusolver-cu12==11.6.1.9
[pip3] nvidia-cusparse-cu12==12.3.1.170
[pip3] nvidia-ml-py==12.560.30
[pip3] nvidia-nccl-cu12==2.21.5
[pip3] nvidia-nvjitlink-cu12==12.4.127
[pip3] nvidia-nvtx-cu12==12.4.127
[pip3] pyzmq==26.2.0
[pip3] torch==2.5.1
[pip3] torchao==0.7.0
[pip3] torchvision==0.20.1
[pip3] transformers==4.48.0
[pip3] triton==3.1.0
[conda] flashinfer                0.1.6+cu121torch2.4          pypi_0    pypi
[conda] numpy                     1.26.4                   pypi_0    pypi
[conda] nvidia-cublas-cu12        12.4.5.8                 pypi_0    pypi
[conda] nvidia-cuda-cupti-cu12    12.4.127                 pypi_0    pypi
[conda] nvidia-cuda-nvrtc-cu12    12.4.127                 pypi_0    pypi
[conda] nvidia-cuda-runtime-cu12  12.4.127                 pypi_0    pypi
[conda] nvidia-cudnn-cu12         9.1.0.70                 pypi_0    pypi
[conda] nvidia-cufft-cu12         11.2.1.3                 pypi_0    pypi
[conda] nvidia-curand-cu12        10.3.5.147               pypi_0    pypi
[conda] nvidia-cusolver-cu12      11.6.1.9                 pypi_0    pypi
[conda] nvidia-cusparse-cu12      12.3.1.170               pypi_0    pypi
[conda] nvidia-ml-py              12.560.30                pypi_0    pypi
[conda] nvidia-nccl-cu12          2.21.5                   pypi_0    pypi
[conda] nvidia-nvjitlink-cu12     12.4.127                 pypi_0    pypi
[conda] nvidia-nvtx-cu12          12.4.127                 pypi_0    pypi
[conda] pyzmq                     26.2.0                   pypi_0    pypi
[conda] torch                     2.5.1                    pypi_0    pypi
[conda] torchao                   0.7.0                    pypi_0    pypi
[conda] torchvision               0.20.1                   pypi_0    pypi
[conda] transformers              4.48.0                   pypi_0    pypi
[conda] triton                    3.1.0                    pypi_0    pypi
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.6.6.post2.dev299+gecf67814
vLLM Build Flags:
CUDA Archs: 5.2 6.0 6.1 7.0 7.2 7.5 8.0 8.6 8.7 9.0+PTX; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0    GPU1    GPU2    GPU3    GPU4    GPU5    GPU6    GPU7    NIC0    NIC1    NIC2    NIC3    NIC4    NIC5    NIC6    NIC7    NIC8    NIC9    NIC10   NIC11   CPU Affinity    NUMA Affinity     GPU NUMA ID
GPU0     X      NV18    NV18    NV18    NV18    NV18    NV18    NV18    PIX     NODE    NODE    NODE    NODE    NODE    NODE    NODE    SYS     SYS     SYS     SYS     0-47,96-143       0               N/A
GPU1    NV18     X      NV18    NV18    NV18    NV18    NV18    NV18    NODE    PIX     NODE    NODE    NODE    NODE    NODE    NODE    SYS     SYS     SYS     SYS     0-47,96-143       0               N/A
GPU2    NV18    NV18     X      NV18    NV18    NV18    NV18    NV18    NODE    NODE    PIX     NODE    NODE    NODE    NODE    NODE    SYS     SYS     SYS     SYS     0-47,96-143       0               N/A
GPU3    NV18    NV18    NV18     X      NV18    NV18    NV18    NV18    NODE    NODE    NODE    NODE    NODE    NODE    NODE    PIX     SYS     SYS     SYS     SYS     0-47,96-143       0               N/A
GPU4    NV18    NV18    NV18    NV18     X      NV18    NV18    NV18    SYS     SYS     SYS     SYS     SYS     SYS     SYS     SYS     PIX     NODE    NODE    NODE    48-95,144-191     1               N/A
GPU5    NV18    NV18    NV18    NV18    NV18     X      NV18    NV18    SYS     SYS     SYS     SYS     SYS     SYS     SYS     SYS     NODE    PIX     NODE    NODE    48-95,144-191     1               N/A
GPU6    NV18    NV18    NV18    NV18    NV18    NV18     X      NV18    SYS     SYS     SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE    PIX     NODE    48-95,144-191     1               N/A
GPU7    NV18    NV18    NV18    NV18    NV18    NV18    NV18     X      SYS     SYS     SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE    NODE    PIX     48-95,144-191     1               N/A
NIC0    PIX     NODE    NODE    NODE    SYS     SYS     SYS     SYS      X      NODE    NODE    NODE    NODE    NODE    NODE    NODE    SYS     SYS     SYS     SYS
NIC1    NODE    PIX     NODE    NODE    SYS     SYS     SYS     SYS     NODE     X      NODE    NODE    NODE    NODE    NODE    NODE    SYS     SYS     SYS     SYS
NIC2    NODE    NODE    PIX     NODE    SYS     SYS     SYS     SYS     NODE    NODE     X      NODE    NODE    NODE    NODE    NODE    SYS     SYS     SYS     SYS
NIC3    NODE    NODE    NODE    NODE    SYS     SYS     SYS     SYS     NODE    NODE    NODE     X      PIX     PXB     PXB     NODE    SYS     SYS     SYS     SYS
NIC4    NODE    NODE    NODE    NODE    SYS     SYS     SYS     SYS     NODE    NODE    NODE    PIX      X      PXB     PXB     NODE    SYS     SYS     SYS     SYS
NIC5    NODE    NODE    NODE    NODE    SYS     SYS     SYS     SYS     NODE    NODE    NODE    PXB     PXB      X      PIX     NODE    SYS     SYS     SYS     SYS
NIC6    NODE    NODE    NODE    NODE    SYS     SYS     SYS     SYS     NODE    NODE    NODE    PXB     PXB     PIX      X      NODE    SYS     SYS     SYS     SYS
NIC7    NODE    NODE    NODE    PIX     SYS     SYS     SYS     SYS     NODE    NODE    NODE    NODE    NODE    NODE    NODE     X      SYS     SYS     SYS     SYS
NIC8    SYS     SYS     SYS     SYS     PIX     NODE    NODE    NODE    SYS     SYS     SYS     SYS     SYS     SYS     SYS     SYS      X      NODE    NODE    NODE
NIC9    SYS     SYS     SYS     SYS     NODE    PIX     NODE    NODE    SYS     SYS     SYS     SYS     SYS     SYS     SYS     SYS     NODE     X      NODE    NODE
NIC10   SYS     SYS     SYS     SYS     NODE    NODE    PIX     NODE    SYS     SYS     SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE     X      NODE
NIC11   SYS     SYS     SYS     SYS     NODE    NODE    NODE    PIX     SYS     SYS     SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE    NODE     X 

Legend:

  X    = Self
  SYS  = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
  NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
  PHB  = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
  PXB  = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
  PIX  = Connection traversing at most a single PCIe bridge
  NV#  = Connection traversing a bonded set of # NVLinks

NIC Legend:

  NIC0: mlx5_0
  NIC1: mlx5_1
  NIC2: mlx5_2
  NIC3: mlx5_3
  NIC4: mlx5_4
  NIC5: mlx5_5
  NIC6: mlx5_6
  NIC7: mlx5_7
  NIC8: mlx5_8
  NIC9: mlx5_9
  NIC10: mlx5_10
  NIC11: mlx5_11

NCCL_DEBUG=INFO
NCCL_IB_HCA=mlx5_0,mlx5_1,mlx5_2,mlx5_7,mlx5_8,mlx5_9,mlx5_10,mlx5_11
PYTORCH_BUILD_NUMBER=0
NCCL_IB_DISABLE=0
NVIDIA_PYTORCH_VERSION=24.03
PYTORCH_BUILD_VERSION=2.3.0a0+40ec155e58
CUDA_CACHE_DISABLE=1
NCCL_VERSION=2.20.5
CUDACXX=/usr/local/cuda/bin/nvcc
CUDA_VERSION=12.4.0.041
NVIDIA_REQUIRE_JETPACK_HOST_MOUNTS=
CUDA_PATH=/usr/local/cuda
PYTORCH_VERSION=2.3.0a0+40ec155e58
NVIDIA_DRIVER_CAPABILITIES=compute,utility,video
TORCH_ALLOW_TF32_CUBLAS_OVERRIDE=1
CUDA_MODULE_LOADING=LAZY
NCCL_NVLS_ENABLE=0
NVIDIA_VISIBLE_DEVICES=GPU-d9b776cd-b371-f675-f545-553ef87d1a1b,GPU-2e6d549d-b5fa-204e-5b5c-441fcddb8d37,GPU-1ef38b97-caeb-9c64-6f10-68cf0f15b7db,GPU-b34a1970-5902-a22c-e03b-2c40784ce17c,GPU-ab34f280-b8dd-cada-8a25-f4b7a57f4d42,GPU-91581554-3c19-3fe8-93fa-0789a405e972,GPU-314a0b69-39da-c1ff-715a-acce32d29a00,GPU-40881086-0988-aabc-da23-d737290ac452
CUBLAS_VERSION=12.4.2.65
CUDA_DRIVER_VERSION=550.54.14
LD_LIBRARY_PATH=/root/miniconda3/envs/sglang/lib/python3.10/site-packages/cv2/../../lib64:/usr/local/cuda/lib64:/usr/local/lib/python3.10/dist-packages/torch/lib:/usr/local/lib/python3.10/dist-packages/torch_tensorrt/lib:/usr/local/cuda/compat/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64
CUDA_HOME=/usr/local/cuda
CUDA_HOME=/usr/local/cuda
NVIDIA_PRODUCT_NAME=PyTorch
NCCL_IB_RETRY_CNT=12
NCCL_IB_QPS_PER_CONNECTION=8
NVIDIA_BUILD_ID=85286408
TORCH_CUDA_ARCH_LIST=5.2 6.0 6.1 7.0 7.2 7.5 8.0 8.6 8.7 9.0+PTX
PYTORCH_HOME=/opt/pytorch/pytorch
NCCL_IB_TIMEOUT=22
NVIDIA_REQUIRE_CUDA=cuda>=9.0
CUDNN_VERSION=9.0.0.306+cuda12.3
NCCL_WORK_FIFO_DEPTH=4194304
NCCL_SOCKET_IFNAME=eth0
NCCL_SOCKET_FAMILY=AF_INET
MAX_JOBS=8
NCCL_CUMEM_ENABLE=0
TORCHINDUCTOR_COMPILE_THREADS=1

Model Input Dumps

No response

🐛 Describe the bug

The issue was observed in the comment posted by @DarkLight1337 #12207 (comment). I've tested the output of different versions of vllm and transformers and have some observations:

1. The output becomes nonsensical after [Model] Upgrade Aria to transformers 4.48 #12203

CEÕ̃ Ã...’ CEOà ÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃÃ...’...’...’...’ tomorrow tomorrow tomorrow tomorrow CEO tomorrow CEO tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow✡ tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrow tomorrowà tomorrow tomorrow tomorrow tomorrow tomorrow��дорÃ

The inference code

from PIL import Image
from transformers import AutoTokenizer
from vllm import LLM, SamplingParams


def main():
    llm = LLM(
        model="rhymes-ai/Aria",
        tokenizer_mode="slow",
        limit_mm_per_prompt={
            "image": 2
        },
        max_num_seqs=2,
        dtype="bfloat16",
    )

    tokenizer = AutoTokenizer.from_pretrained(
        "rhymes-ai/Aria", use_fast=False
    )

    messages = [
        {
            "role": "user",
            "content": [
                {
                    "type": "text",
                    "text": "What is in the image?"
                },
                {"type": "image"},
            ],
        }
    ]

    message = tokenizer.apply_chat_template(messages, add_generation_prompt=True)

    outputs = llm.generate(
        {
            "prompt_token_ids": message,
            "multi_modal_data": {
                "image": [
                    Image.open("/root/.cache/vllm/assets/vllm_public_assets/cherry_blossom.jpg"),
                ],
            },
        },
        sampling_params=SamplingParams(max_tokens=200, top_k=1, stop=["<|im_end|>"]),
    )

    for o in outputs:
        generated_tokens = o.outputs[0].token_ids
        print(tokenizer.decode(generated_tokens))


if __name__ == "__main__":
    main()

2. vLLM 0.6.6, transformers 4.45.0 with an older revision of Aria hf repo produce the correct output

The image shows a beautiful scene with cherry blossoms in full bloom. The blossoms are pink and cover the branches of the trees, creating a canopy of flowers. In the background, there is a tall, white tower with a spherical structure at the top, which appears to be the Tokyo Skytree, a famous landmark in Japan. The sky is clear and blue, adding to the serene and picturesque quality of the image.<|im_end|>

The inference code

from PIL import Image
from transformers import AutoTokenizer
from vllm import LLM, SamplingParams

revision = "4844f0b5ff678e768236889df5accbe4967ec845"

def main():
    llm = LLM(
        model="rhymes-ai/Aria",
        revision=revision,
        tokenizer_mode="slow",
        dtype="bfloat16",
        trust_remote_code=True,
    )

    tokenizer = AutoTokenizer.from_pretrained(
        "rhymes-ai/Aria", revision=revision, trust_remote_code=True, use_fast=False
    )

    messages = [
        {
            "role": "user",
            "content": [
                {
                    "type": "text",
                    "text": "What is in the image?"
                },
                {"type": "image"},
            ],
        }
    ]

    message = tokenizer.apply_chat_template(messages, add_generation_prompt=True)

    outputs = llm.generate(
        {
            "prompt_token_ids": message,
            "multi_modal_data": {
                "image": [
                    Image.open("/root/.cache/vllm/assets/vllm_public_assets/cherry_blossom.jpg"),
                ],
            },
        },
        sampling_params=SamplingParams(max_tokens=200, top_k=1, stop=["<|im_end|>"]),
    )

    for o in outputs:
        generated_tokens = o.outputs[0].token_ids
        print(tokenizer.decode(generated_tokens))


if __name__ == "__main__":
    main()

3. The inference output of transformers 4.48 is correct

The image features a prominent structure that appears to be a tall communication tower set against a clear blue sky. The tower is surrounded by lush branches with vibrant pink blossoms, likely cherry blossoms, adding a beautiful and serene atmosphere to the scene. The contrast between the architectural element and the natural elements creates a visually striking and harmonious composition. The image captures the essence of a peaceful setting blending technology with nature. <|im_end|>

The inference code

import torch
from PIL import Image

from transformers import AriaProcessor, AriaForConditionalGeneration


model_id_or_path = "rhymes-ai/Aria"
model = AriaForConditionalGeneration.from_pretrained(
    model_id_or_path, device_map="auto", torch_dtype=torch.bfloat16
)

processor = AriaProcessor.from_pretrained(model_id_or_path)

image = Image.open("/root/.cache/vllm/assets/vllm_public_assets/cherry_blossom.jpg")

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image"},
            {"text": "what is the image?", "type": "text"},
        ],
    }
]

text = processor.apply_chat_template(messages, add_generation_prompt=True)
inputs = processor(text=text, images=image, return_tensors="pt")
inputs['pixel_values'] = inputs['pixel_values'].to(torch.bfloat16)
inputs.to(model.device)

output = model.generate(
    **inputs,
    max_new_tokens=200,
    stop_strings=["<|im_end|>"],
    tokenizer=processor.tokenizer,
    do_sample=True,
    temperature=0.9,
)
output_ids = output[0][inputs["input_ids"].shape[1]:]
response = processor.decode(output_ids, skip_special_tokens=True)
print(response)

cc @DarkLight1337 @Isotr0py

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[Isotr0py]

I remember this model is broken earlier than #12203, because the nonsensical outputs were firstly observed during the development of #11632, and the model has broken on main branch at that time.

[DarkLight1337]

I found that I can't load the model in transformers v4.48 if trust_remote_code=True. @xffxff can you update https://huggingface.co/rhymes-ai/Aria/blob/main/config.json so that auto_map points to the classes defined by transformers instead of the model repo? I think you just need to remove the modeling_aria. prefix.

[xffxff]

I found that I can't load the model in transformers v4.48 if trust_remote_code=True. @xffxff can you update https://huggingface.co/rhymes-ai/Aria/blob/main/config.json so that auto_map points to the classes defined by transformers instead of the model repo? I think you just need to remove the modeling_aria. prefix.

I've updated the hf repo. Can you give it a try? @DarkLight1337

[DarkLight1337]

Still doesn't work. Maybe we should just remove auto_map? I assume those classes are already registered automatically inside transformers repo, like here: https://github.com/huggingface/transformers/blob/5615a393691c81e00251e420c73e4d04c6fe22e5/src/transformers/models/auto/modeling_auto.py#L38

[xffxff]

Still doesn't work. Maybe we should just remove auto_map? I assume those models are already registered automatically inside transformers repo.

I've removed automap. Can you share your code so I can check it in my env if it still doesn't work

[DarkLight1337]

I've removed automap. Can you share your code so I can check it in my env if it still doesn't work

It works now, thanks!

[xffxff]

One bug I've found is that the vision tower's output is a tensor with all zero values. That is because the aria model uses Idefics3VisionTransformer without the post_norm layer. So we can not just use ``Idefics3VisionTransformer` as the vision tower directly. Below is the original Aria impl in vLLM

vllm/vllm/model_executor/models/aria.py

Lines 57 to 70 in b1d9205

	class AriaVisionTransformer(Idefics2VisionTransformer):
	"""
	AriaVisionTransformer is a modified version of Idefics2VisionTransformer
	that replaces the post-layernorm with an identity layer.
	"""
	def __init__(
	self,
	config: AriaVisionConfig,
	quant_config: Optional[QuantizationConfig] = None,
	prefix: str = "",
	) -> None:
	super().__init__(config, quant_config, prefix)
	self.post_layernorm = nn.Identity()

The current checkpoint in hf repo after https://huggingface.co/rhymes-ai/Aria/discussions/11 includes the weights for post_norm layer , where all the weights to be zeros.

The transformers impl of Aria avoids this issue by using the output before the post_norm as the final vision tower output. it sets output_hidden_states=True and gets the hidden states at vision_feature_layer, which is the output before post_norm

https://github.com/huggingface/transformers/blob/f4f33a20a23aa90f3510280e34592b2784d48ebe/src/transformers/models/aria/modeling_aria.py#L1417-L1425

Sadly, even after fixing the issue, the output is still nonsensical 😂