Code

sync : llama.cpp#1051

Code

Merged

ggerganov merged 19 commits intomasterggerganov/ggml:masterfrom sync-llama.cpp-24-12-17-0ggerganov/ggml:sync-llama.cpp-24-12-17-0Copy head branch name to clipboardDec 17, 2024

+1,682-616

Commits

Commits on Dec 17, 2024

• vulkan: disable spirv-opt for coopmat shaders (llama/10763)

  Show description for f8f71ea
  

  jeffbolznv

  authored and

  ggerganov

  committed
  f8f71ea

• CUDA: rename macros to avoid conflicts with WinAPI (llama/10736)

  Show description for 259296c
  

  aendk

  authored and

  ggerganov

  committed
  259296c

• vulkan: dynamic subgroup size for the remaining k quants (llama/10745)

  Show description for 8f8e2c8
  

  netrunnereve

  authored and

  ggerganov

  committed
  8f8e2c8

• vulkan: request round-to-even for fp16 in im2col/rope_head (llama/10767)

  Show description for d9b9571
  

  jeffbolznv

  authored and

  ggerganov

  committed
  d9b9571

• ggml: load all backends from a user-provided search path (llama/10699)

  Show description for 4847921
  

  2 peopleauthored and

  ggerganov

  committed
  4847921

• Vulkan: Add VK_EXT_subgroup_size_control support to ensure full subgroups for coopmats (llama/10721)

  Show description for feec94d
  

  0cc4m

  authored and

  ggerganov

  committed
  feec94d

• Vulkan: Use improved q4_k and q5_k dequant code in dequant shaders (llama/10798)

  

  0cc4m

  authored and

  ggerganov

  committed
  de05eb6

• remove CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS (llama/10797)

  Show description for 68c517e
  

  slaren

  authored and

  ggerganov

  committed
  68c517e

• CUDA: faster non-contiguous concat (llama/10760)

  Show description for bfa090f
  

  2 peopleauthored and

  ggerganov

  committed
  bfa090f

• ggml : Fix compilation issues on ARM platform when building without fp16 (llama/10811)

  

  kkontny

  authored and

  ggerganov

  committed
  72e030c

• SYCL: Reduce most of the compiler warnings (llama/10748)

  Show description for e6cdbe9
  

  2 peopleauthored and

  ggerganov

  committed
  e6cdbe9

• vulkan: small mul_mat_vec optimizations (llama/10665)

  Show description for 793b3fd
  

  netrunnereve

  authored and

  ggerganov

  committed
  793b3fd

• Fix crash caused by ggml_backend_load_all when launching on Android Activity (llama/10812)

  Show description for 42609f8
  

  2 peopleauthored and

  ggerganov

  committed
  42609f8

• Introducing experimental OpenCL backend with support for Qualcomm Adreno GPUs (llama/10693)

  Show description for b36f93a
  

  6 peopleauthored and

  ggerganov

  committed
  b36f93a

• llama : add Qwen2VL support + multimodal RoPE (llama/10361)

  Show description for 118c41d
  

  HimariO

  and

  ggerganov

  committed
  118c41d

• rwkv6: add wkv6 support for Vulkan backend (llama/10829)

  Show description for 3a02d27
  

  2 peopleauthored and

  ggerganov

  committed
  3a02d27

• vulkan: bugfixes for small subgroup size systems + llvmpipe test (llama/10809)

  Show description for 00d6f1b
  

  netrunnereve

  authored and

  ggerganov

  committed
  00d6f1b

• ggml : update ggml_backend_cpu_device_supports_op (llama/10867)

  Show description for 6690443
  ggerganov

  committed
  6690443

• sync : llama.cpp

  Show description for d7ae184
  ggerganov

  committed
  d7ae184