Code

b2915#125

Code

Merged

Nexesenex merged 13 commits intoNexesenex:downstreamNexesenex/croco.cpp:downstreamfrom ggerganov:masterggerganov/llama.cpp:masterCopy head branch name to clipboardMay 17, 2024

+1,519-400

Commits

Commits on May 16, 2024

• Revert "server bench: fix bench not waiting for model load (#7284)" (#7334)

  Show description for 24ecb58
  phymbert

  authored
  24ecb58

  Copy full SHA for 24ecb58

Commits on May 17, 2024

• [Server] Added --verbose option to README [no ci] (#7335)

  reuank

  authored
  9c4fdcb

  Copy full SHA for 9c4fdcb

• ggml : rewrite silu and softmax for cpu (#7154)

  Show description for 934266c
  jart

  authored
  934266c

  Copy full SHA for 934266c

• server : add support for the RPC backend (#7305)

  Show description for ee94172
  rgerganov

  authored
  ee94172

  Copy full SHA for ee94172

• convert : fix Qwen/Qwen-7b conversion (#7308)

  amd-lalithnc

  authored
  e18bc6a

  Copy full SHA for e18bc6a

• ggml-quants, llama : removed excess checks (#7274)

  GermanAizek

  authored
  359cbe3

  Copy full SHA for 359cbe3

• tokenization: add warning for double BOS (#7332)

  JohannesGaessler

  authored
  29c60d8

  Copy full SHA for 29c60d8

• llama : use n_embd_head_v when reshaping kqv (#7327)

  Show description for 27b0406
  

  fairydreaming

  and

  sszymczy

  authored
  27b0406

  Copy full SHA for 27b0406

• py : convert-hf-to-gguf-update improvements (#7340)

  Show description for d273c14
  akx

  authored
  d273c14

  Copy full SHA for d273c14

• Added a single test function script and fix debug-test.sh to be more robust (#7279)

  Show description for 51e9d02
  mofosyne

  authored
  51e9d02

  Copy full SHA for 51e9d02

• rpc : set SO_REUSEADDR for the server socket (#7320)

  Show description for f4bd8b3
  rgerganov

  authored
  f4bd8b3

  Copy full SHA for f4bd8b3

• ROCm: use native CMake HIP support (#5966)

  Show description for 82ca83d
  GZGavinZhao

  authored
  82ca83d

  Copy full SHA for 82ca83d

• CUDA: faster large batch FA without tensor cores (#7314)

  JohannesGaessler

  authored
  0fc1e82

  Copy full SHA for 0fc1e82