Code

sync : llama.cpp#1184

Code

Merged

ggerganov merged 27 commits intomasterggml-org/ggml:masterfrom

sync-llama-25-04-08ggml-org/ggml:sync-llama-25-04-08Copy head branch name to clipboard

Apr 8, 2025

Commits

Commits on Apr 8, 2025

• llama : add option to override model tensor buffers (llama/11397)

  Show description for 53d7f83
  

  slaren

  authored and

  ggerganov

  committed
  53d7f83

  Copy full SHA for 53d7f83

• Vulkan: Fix mmq int dot float cache size (llama/12722)

  

  0cc4m

  authored and

  ggerganov

  committed
  d3b2ef6

  Copy full SHA for d3b2ef6

• vulkan: Implement grouped query attention in the coopmat2 FA shader (llama/12559)

  Show description for 389b652
  

  jeffbolznv

  authored and

  ggerganov

  committed
  389b652

  Copy full SHA for 389b652

• cmake: remove caching from vulkan coopmat checks (llama/12719)

  

  bandoti

  authored and

  ggerganov

  committed
  b1cb823

  Copy full SHA for b1cb823

• vulkan: Implement split_k for coopmat2 flash attention. (llama/12627)

  Show description for 493c545
  

  jeffbolznv

  authored and

  ggerganov

  committed
  493c545

  Copy full SHA for 493c545

• opencl: use `max_alloc_size` in backend ctx instead of querying again (llama/12705)

  

  lhez

  authored and

  ggerganov

  committed
  cd2abee

  Copy full SHA for cd2abee

• CANN: Fix failed test cases (llama/12708)

  Show description for 2de3d0e
  

  hipudding

  authored and

  ggerganov

  committed
  2de3d0e

  Copy full SHA for 2de3d0e

• Simplify and improve CUDA graphs through use of indirect copy pointers (llama/9017)

  Show description for 1fc73a4
  

  2 peopleauthored and

  ggerganov

  committed
  1fc73a4

  Copy full SHA for 1fc73a4

• CANN: Support operator SIN COS ARGMAX (llama/12709)

  Show description for 9cf60e4
  

  2 peopleauthored and

  ggerganov

  committed
  9cf60e4

  Copy full SHA for 9cf60e4

• fix MUSA compiler warning (llama/12704)

  Show description for 0e538c4
  

  A3shTnT

  authored and

  ggerganov

  committed
  0e538c4

  Copy full SHA for 0e538c4

• vulkan: Fix missing cmake logic for dot product extension (llama/12721)

  

  jeffbolznv

  authored and

  ggerganov

  committed
  cdb685e

  Copy full SHA for cdb685e

• CUDA: Prefer vector flash decoding kernel for Gemma models (llama/12738)

  Show description for 9fa053e
  

  2 peopleauthored and

  ggerganov

  committed
  9fa053e

  Copy full SHA for 9fa053e

• vulkan: set cmake minimum and project name in vulkan-shaders (llama/12744)

  

  jeffbolznv

  authored and

  ggerganov

  committed
  a17eace

  Copy full SHA for a17eace

• vulkan: Hybrid waitForFences/getFenceStatus to reduce fence latency (llama/12630)

  Show description for 107f582
  

  jeffbolznv

  authored and

  ggerganov

  committed
  107f582

  Copy full SHA for 107f582

• cmake: fix ggml-shaders-gen compiler paths containing spaces (llama/12747)

  Show description for 21e404f
  

  hydroo

  authored and

  ggerganov

  committed
  21e404f

  Copy full SHA for 21e404f

• sycl: allow ggml-sycl configuration and compilation using Visual Studio project/solution (llama/12625)

  

  s-Nick

  authored and

  ggerganov

  committed
  babf217

  Copy full SHA for babf217

• Vulkan: Tune Vulkan mmq int dot shader for performance (llama/12767)

  

  0cc4m

  authored and

  ggerganov

  committed
  1a67444

  Copy full SHA for 1a67444

• vulkan: Use unclamped loads for flash attention mask (llama/12720)

  Show description for ff13fab
  

  jeffbolznv

  authored and

  ggerganov

  committed
  ff13fab

  Copy full SHA for ff13fab

• vulkan: fix NaN issue in flash attention shader (llama/12776)

  Show description for 8d5fa1e
  

  jeffbolznv

  authored and

  ggerganov

  committed
  8d5fa1e

  Copy full SHA for 8d5fa1e

• musa: fix compilation warnings in mp_22/31 (llama/12780)

  Show description for 631561f
  

  yeahdongcn

  authored and

  ggerganov

  committed
  631561f

  Copy full SHA for 631561f

• CANN: Refactor to reduce duplicate code (llama/12731)

  Show description for 34474b3
  

  hipudding

  authored and

  ggerganov

  committed
  34474b3

  Copy full SHA for 34474b3

• CANN: fix typo in ggml-cann (llama/12733)

  

  zhouwg

  authored and

  ggerganov

  committed
  386bf69

  Copy full SHA for 386bf69

• sycl: remove redundant memcopy in function ggml_backend_sycl_buffer_set_tensor (llama/12734)

  

  zhouwg

  authored and

  ggerganov

  committed
  7eeb2aa

  Copy full SHA for 7eeb2aa

• cuda : fix HIP and MUSA BF16 (llama/0)

  Show description for b0e5715
  ggerganov

  committed
  b0e5715

  Copy full SHA for b0e5715

• opencl: better identify Adreno GPU (llama/12760)

  

  lhez

  authored and

  ggerganov

  committed
  26e2e53

  Copy full SHA for 26e2e53

• Revert "sycl:remove redundant memcopy in function ggml_backend_sycl_buffer_set_tensor" (llama/12812)

  Show description for 508beb7
  

  NeoZhangJianyu

  authored and

  ggerganov

  committed
  508beb7

  Copy full SHA for 508beb7

• sync : llama.cpp

  Show description for 54f2309
  ggerganov

  committed
  54f2309

  Copy full SHA for 54f2309