Code

sync : llama.cpp#1134

Code

Merged

ggerganov merged 14 commits intomasterggml-org/ggml:masterfrom sync-llama-25-03-07ggml-org/ggml:sync-llama-25-03-07Copy head branch name to clipboardMar 7, 2025

Commits

Commits on Mar 7, 2025

• ggml : fix GGMLMetalClass ODR (llama/12200)

  Show description for a60363b
  

  pminev

  authored and

  ggerganov

  committed
  a60363b

  Copy full SHA for a60363b

• SYCL: Disable f16 Unary OPs as not supported by the kernels (llama/12201)

  

  qnixsynapse

  authored and

  ggerganov

  committed
  8c382a5

  Copy full SHA for 8c382a5

• ggml-cpu: Faster IQ1 mul_mat_vec on AVX2 using BMI2 instructions (llama/12154)

  Show description for 5f043dd
  

  remyoudompheng

  authored and

  ggerganov

  committed
  5f043dd

  Copy full SHA for 5f043dd

• opencl : fix profile-related errors (llama/12095)

  Show description for a6b96bc
  

  2 peopleauthored and

  ggerganov

  committed
  a6b96bc

  Copy full SHA for a6b96bc

• opencl : fix ulong kernel args were set from int variables (llama/12174)

  Show description for 4fbd566
  

  linehill

  authored and

  ggerganov

  committed
  4fbd566

  Copy full SHA for 4fbd566

• opencl : fix buffer alignment (llama/12197)

  Show description for e128693
  

  linehill

  authored and

  ggerganov

  committed
  e128693

  Copy full SHA for e128693

• HIP/CUDA: set the paramerter value in maintain_cuda_graph instead of replaceing it. (llama/12209)

  Show description for a5020df
  

  IMbackK

  authored and

  ggerganov

  committed
  a5020df

  Copy full SHA for a5020df

• CUDA: fix FA logic for PTX 7.0 and CC >= 7.5 (llama/12222)

  

  JohannesGaessler

  authored and

  ggerganov

  committed
  0f13ede

  Copy full SHA for 0f13ede

• cmake : fix undefined reference errors for std::filesystem in ggml (#12092) (llama/12094)

  Show description for 1993982
  

  2 peopleauthored and

  ggerganov

  committed
  1993982

  Copy full SHA for 1993982

• opencl: Noncontiguous norm, rms_norm, disable fp16 for some ops (llama/12217)

  Show description for 381c382
  

  lhez

  authored and

  ggerganov

  committed
  381c382

  Copy full SHA for 381c382

• metal : fix default.metallib build (llama/12224)

  Show description for 1631cc0
  

  danbev

  authored and

  ggerganov

  committed
  1631cc0

  Copy full SHA for 1631cc0

• metal : simplify kernel arguments using a struct (#3229) (llama/12194)

  Show description for 23e4afd
  

  2 peopleauthored and

  ggerganov

  committed
  23e4afd

  Copy full SHA for 23e4afd

• ggml-cpu: faster AVX2 variant for IQ1_M (llama/12216)

  

  remyoudompheng

  authored and

  ggerganov

  committed
  61022c3

  Copy full SHA for 61022c3

• sync : llama.cpp

  Show description for 90876ef
  ggerganov

  committed
  90876ef

  Copy full SHA for 90876ef