Code

sync : llama.cpp#1070

Code

Merged

ggerganov merged 23 commits intomasterggerganov/ggml:masterfrom sync-llama.cpp-25-01-14ggerganov/ggml:sync-llama.cpp-25-01-14Copy head branch name to clipboardJan 14, 2025

+12,498-2,083

Commits

Commits on Jan 14, 2025

• fix: Vulkan shader gen binary path (llama/11037)

  

  giladgd

  authored and

  ggerganov

  committed
  b19d9cc

  Copy full SHA for b19d9cc

• Support for models with non-512-aligned tensors over RPC. (llama/11047)

  Show description for dc5f0bc
  

  2 peopleauthored and

  ggerganov

  committed
  dc5f0bc

  Copy full SHA for dc5f0bc

• Vulkan: Add device-specific blacklist for coopmat for the AMD proprietary driver (llama/11074)

  Show description for 81fde73
  

  0cc4m

  authored and

  ggerganov

  committed
  81fde73

  Copy full SHA for 81fde73

• CUDA: add BF16 support (llama/11093)

  Show description for 5514692
  

  JohannesGaessler

  authored and

  ggerganov

  committed
  5514692

  Copy full SHA for 5514692

• SYCL: Use get_multi_ptr instead of deprecated get_pointer in wkv6 (llama/11087)

  Show description for 20a72f1
  

  qnixsynapse

  authored and

  ggerganov

  committed
  20a72f1

  Copy full SHA for 20a72f1

• rpc : code cleanup (llama/11107)

  Show description for ef9e2eb
  

  rgerganov

  authored and

  ggerganov

  committed
  ef9e2eb

  Copy full SHA for ef9e2eb

• ggml-backend : only offload from host buffers (llama/11120)

  

  slaren

  authored and

  ggerganov

  committed
  4630db8

  Copy full SHA for 4630db8

• ggml-backend : only offload from host buffers (fix) (llama/11124)

  

  slaren

  authored and

  ggerganov

  committed
  7150212

  Copy full SHA for 7150212

• GGUF: C++ refactor, backend support, misc fixes (llama/11030)

  Show description for bacc721
  

  JohannesGaessler

  authored and

  ggerganov

  committed
  bacc721

  Copy full SHA for bacc721

• fix: Vulkan shader gen binary path when Cross-compiling (llama/11096)

  Show description for 6a87a43
  

  ag2s20150909

  authored and

  ggerganov

  committed
  6a87a43

  Copy full SHA for 6a87a43

• Disable GL_KHR_cooperative_matrix Vulkan extension if not available. (llama/11117)

  Show description for 454cfa3
  

  mbaudier

  authored and

  ggerganov

  committed
  454cfa3

  Copy full SHA for 454cfa3

• llamafile : ppc64le MMA INT8 implementation (llama/10912)

  Show description for 7de0ddc
  

  amritahs-ibm

  authored and

  ggerganov

  committed
  7de0ddc

  Copy full SHA for 7de0ddc

• fix: add missing msg in static_assert (llama/11143)

  Show description for d50565e
  

  hydai

  authored and

  ggerganov

  committed
  d50565e

  Copy full SHA for d50565e

• SYCL: Refactor ggml_sycl_compute_forward (llama/11121)

  Show description for 620b188
  

  qnixsynapse

  authored and

  ggerganov

  committed
  620b188

  Copy full SHA for 620b188

• llama: add support for QRWKV6 model architecture (llama/11001)

  Show description for 17b0a11
  

  3 peoplecommitted
  17b0a11

  Copy full SHA for 17b0a11

• Vulkan: Fix float16 use on devices without float16 support + fix subgroup_size_control validation error (llama/11161)

  Show description for 67b8cc9
  

  0cc4m

  authored and

  ggerganov

  committed
  67b8cc9

  Copy full SHA for 67b8cc9

• ggml : do not define GGML_USE_CUDA when building with GGML_BACKEND_DL (llama/11211)

  Show description for e819be6
  

  rgerganov

  authored and

  ggerganov

  committed
  e819be6

  Copy full SHA for e819be6

• cuda : CUDA Graph Compute Function Refactor (precursor for performance improvements) (llama/11042)

  Show description for aad9107
  

  2 peopleauthored and

  ggerganov

  committed
  aad9107

  Copy full SHA for aad9107

• sync : llama.cpp

  Show description for 23fbf2f
  ggerganov

  committed
  23fbf2f

  Copy full SHA for 23fbf2f

• scripts : sync opencl

  ggerganov

  committed
  f6cbb38

  Copy full SHA for f6cbb38

• ggml : add opencl backend (skip) (llama/10693)

  Show description for ced5e67
  

  6 peopleauthored and

  ggerganov

  committed
  ced5e67

  Copy full SHA for ced5e67

• scripts : sync gguf

  ggerganov

  committed
  18e8b10

  Copy full SHA for 18e8b10

• GGUF: C++ refactor, backend support, misc fixes (skip) (llama/11030)

  Show description for daca9a1
  

  JohannesGaessler

  authored and

  ggerganov

  committed
  daca9a1

  Copy full SHA for daca9a1