Code

[pull] master from ggerganov:master#171

Code

Closed

pull[bot] wants to merge 43 commits intoteleprint-me:masterteleprint-me/llama.cpp:masterfrom ggerganov:masterggerganov/llama.cpp:masterCopy head branch name to clipboard

+6,828-870

Commits

Commits on Jan 26, 2025

• readme : update hot topics

  ggerganov

  authored
  2cc9b8c

  Copy full SHA for 2cc9b8c

• rpc: fix register position (#11424)

  Show description for 1d8ee06
  thxCode

  authored
  1d8ee06

  Copy full SHA for 1d8ee06

• cmake: add ggml find package (#11369)

  Show description for 19f6518
  

  bandoti

  and

  ggerganov

  authored
  19f6518

  Copy full SHA for 19f6518

• docker: add missing vulkan library to base layer and update to 24.04 (#11422)

  Show description for 6f53d8a
  rare-magma

  authored
  6f53d8a

  Copy full SHA for 6f53d8a

• metal : use residency sets (#11427)

  Show description for 178a7eb
  ggerganov

  authored
  178a7eb

  Copy full SHA for 178a7eb

• docker : fix ARM build and Vulkan build (#11434)

  Show description for caf773f
  ngxson

  authored
  caf773f

  Copy full SHA for caf773f

Commits on Jan 27, 2025

• metal: Handle null returned from MTLCreateSystemDefaultDevice() (#11441)

  Show description for acd38ef
  booxter

  authored
  acd38ef

  Copy full SHA for acd38ef

• llama: refactor llama_decode_impl (#11381)

  JohannesGaessler

  authored
  df984e0

  Copy full SHA for df984e0

• llama : minor fixes for up llama load model speed (#11448)

  Show description for a5203b4
  

  3 peopleauthored
  a5203b4

  Copy full SHA for a5203b4

• AMD: parse the architecture as supplied by gcnArchName (#11244)

  Show description for d6d24cd
  Haus1

  authored
  d6d24cd

  Copy full SHA for d6d24cd

• Add new hf protocol for ollama (#11449)

  Show description for a4417dd
  ericcurtin

  authored
  a4417dd

  Copy full SHA for a4417dd

Commits on Jan 28, 2025

• Handle missing model in CLI parameters for llama-run (#11399)

  Show description for 2b8525d
  engelmi

  authored
  2b8525d

  Copy full SHA for 2b8525d

• SYCL : SOFTMAX F16 mask support and other fixes (#11261)

  Show description for 6e84b0a
  qnixsynapse

  authored
  6e84b0a

  Copy full SHA for 6e84b0a

• docker: add perplexity and bench commands to full image (#11438)

  Show description for f643120
  rare-magma

  authored
  f643120

  Copy full SHA for f643120

• cmake : don't fail on GGML_CPU=OFF (#11457)

  someone13574

  authored
  4bf3119

  Copy full SHA for 4bf3119

• docker: allow installing pip packages system-wide (#11437)

  Show description for d7d1ecc
  rare-magma

  authored
  d7d1ecc

  Copy full SHA for d7d1ecc

• Add github protocol pulling and http:// (#11465)

  Show description for 7fee288
  ericcurtin

  authored
  7fee288

  Copy full SHA for 7fee288

• HIP: Only call rocblas_initialize on rocblas versions with the multiple instantation bug (#11080)

  Show description for cae9fb4
  sARY77

  authored
  cae9fb4

  Copy full SHA for cae9fb4

• HIP: Supress transformation warning in softmax.cu

  Show description for be5ef79
  IMbackK

  authored
  be5ef79

  Copy full SHA for be5ef79

• ci : fix build CPU arm64 (#11472)

  Show description for d0c0804
  ngxson

  authored
  d0c0804

  Copy full SHA for d0c0804

• server : Fixed wrong function name in llamacpp server unit test (#11473)

  Show description for cf8cc85
  peidaqi

  authored
  cf8cc85

  Copy full SHA for cf8cc85

• cmake: add hints for locating ggml on Windows using Llama find-package (#11466)

  Emreerdog

  authored
  794fe23

  Copy full SHA for 794fe23

Commits on Jan 29, 2025

• llama: fix missing k_cache store for rwkv6qwen2 (#11445)

  Show description for 325afb3
  MollySophia

  authored
  325afb3

  Copy full SHA for 325afb3

• embedding : enable --no-warmup option (#11475)

  Show description for b636228
  danbev

  authored
  b636228

  Copy full SHA for b636228

• ggml-cpu : fix ggml_graph_compute_thread did not terminate on abort. (ggml/1065)

  Show description for d2e518e
  

  2 peopleauthored and

  ggerganov

  committed
  d2e518e

  Copy full SHA for d2e518e

• ggml : add option to not print stack on abort (ggml/1081)

  Show description for 1a0e87d
  

  2 peopleauthored and

  ggerganov

  committed
  1a0e87d

  Copy full SHA for 1a0e87d

• sync : ggml

  ggerganov

  committed
  8158577

  Copy full SHA for 8158577

• Parse https://ollama.com/library/ syntax (#11480)

  Show description for f0d4b29
  ericcurtin

  authored
  f0d4b29

  Copy full SHA for f0d4b29

• vulkan: Catch pipeline creation failure and print an error message (#11436)

  Show description for 2711d02
  jeffbolznv

  authored
  2711d02

  Copy full SHA for 2711d02

• server : update auto gen files comments [no ci] (#11484)

  Show description for e51c47b
  danbev

  authored
  e51c47b

  Copy full SHA for e51c47b

• vulkan: implement initial support for IQ2 and IQ3 quantizations (#11360)

  Show description for 66ee4f2
  

  remyoudompheng

  and

  jeffbolznv

  authored
  66ee4f2

  Copy full SHA for 66ee4f2

• server : add /apply-template endpoint for additional use cases of Minja functionality (#11489)

  Show description for eb7cf15
  pnb

  authored
  eb7cf15

  Copy full SHA for eb7cf15

Commits on Jan 30, 2025

• server : update json snippets in README.md [no ci] (#11492)

  Show description for e044976
  danbev

  authored
  e044976

  Copy full SHA for e044976

• readme : reference examples relative links (#11505)

  guspan-tanadi

  authored
  7919256

  Copy full SHA for 7919256

• server : (docs) added response format for /apply-template [no ci] (#11503)

  isaac-mcfadyen

  authored
  496e5bf

  Copy full SHA for 496e5bf

• server : use lambda instead of std::bind (#11507)

  Show description for 4314e56
  danbev

  authored
  4314e56

  Copy full SHA for 4314e56

• vocab : correctly identify LF token for GPT-2 style BPE tokenizer (#11496)

  mgroeber9110

  authored
  ffd0821

  Copy full SHA for ffd0821

• sync: minja (#11499)

  ochafik

  authored
  3d804de

  Copy full SHA for 3d804de

• CUDA/HIP: add warp_size to cuda_device_info

  IMbackK

  committed
  c300e68

  Copy full SHA for c300e68

• HIP: Prepare reduction operators for wave 64

  IMbackK

  committed
  6af1ca4

  Copy full SHA for 6af1ca4

• HIP: require at least HIP 5.5

  IMbackK

  committed
  27d135c

  Copy full SHA for 27d135c

• Tool call support (generic + native for Llama, Functionary, Hermes, Mistral, Firefunction, DeepSeek) w/ lazy grammars (#9639)

  Show description for 8b576b6
  

  3 peopleauthored
  8b576b6

  Copy full SHA for 8b576b6

• ci: ccache for all github worfklows (#11516)

  ochafik

  authored
  553f1e4

  Copy full SHA for 553f1e4