Code

[pull] master from ggerganov:master#165

Code

Closed

pull[bot] wants to merge 72 commits intoteleprint-me:masterteleprint-me/llama.cpp:masterfrom ggerganov:masterggerganov/llama.cpp:masterCopy head branch name to clipboard

+13,978-9,147

Commits

Commits on Jan 7, 2025

• SYCL: Use get_multi_ptr instead of deprecated get_pointer in wkv6 (#11087)

  Show description for c0d6f79
  qnixsynapse

  authored
  c0d6f79

  Copy full SHA for c0d6f79

• rpc : code cleanup (#11107)

  Show description for a4dd490
  rgerganov

  authored
  a4dd490

  Copy full SHA for a4dd490

• ggml-backend : only offload from host buffers (#11120)

  slaren

  authored
  a3d50bc

  Copy full SHA for a3d50bc

• ggml-backend : only offload from host buffers (fix) (#11124)

  slaren

  authored
  017cc5f

  Copy full SHA for 017cc5f

• GGUF: C++ refactor, backend support, misc fixes (#11030)

  Show description for 53ff6b9
  JohannesGaessler

  authored
  53ff6b9

  Copy full SHA for 53ff6b9

Commits on Jan 8, 2025

• fix: Vulkan shader gen binary path when Cross-compiling (#11096)

  Show description for bec2183
  ag2s20150909

  authored
  bec2183

  Copy full SHA for bec2183

• Disable GL_KHR_cooperative_matrix Vulkan extension if not available. (#11117)

  Show description for 02f0430
  mbaudier

  authored
  02f0430

  Copy full SHA for 02f0430

• ci : fix cmake option (#11125)

  ggerganov

  authored
  0d52a69

  Copy full SHA for 0d52a69

• llamafile : ppc64le MMA INT8 implementation (#10912)

  Show description for 8cef75c
  amritahs-ibm

  authored
  8cef75c

  Copy full SHA for 8cef75c

• arg : option to exclude arguments from specific examples (#11136)

  Show description for a3c1232
  ggerganov

  authored
  a3c1232

  Copy full SHA for a3c1232

• ci : pin dependency to specific version (#11137)

  Show description for 80ccf5d
  ngxson

  authored
  80ccf5d

  Copy full SHA for 80ccf5d

• ggml : allow loading backend with env variable (ggml/1059)

  Show description for c792dcf
  

  rgerganov

  authored and

  ggerganov

  committed
  c792dcf

  Copy full SHA for c792dcf

• sync : ggml

  ggerganov

  committed
  99a3755

  Copy full SHA for 99a3755

• llama : avoid hardcoded QK_K (#11061)

  Show description for c07d437
  ggerganov

  authored
  c07d437

  Copy full SHA for c07d437

• lora : improve compat with mergekit-extract-lora (#11131)

  Show description for 4d2b3d8
  ngxson

  authored
  4d2b3d8

  Copy full SHA for 4d2b3d8

• ci : use actions from ggml-org (#11140)

  ngxson

  authored
  f7cd133

  Copy full SHA for f7cd133

• Enhance user input handling for llama-run (#11138)

  Show description for 1bf839b
  ericcurtin

  authored
  1bf839b

  Copy full SHA for 1bf839b

• gguf-py : move scripts directory (#11116)

  Show description for 8a1d9c2
  VJHack

  authored
  8a1d9c2

  Copy full SHA for 8a1d9c2

• fix: add missing msg in static_assert (#11143)

  Show description for 8d59d91
  hydai

  authored
  8d59d91

  Copy full SHA for 8d59d91

Commits on Jan 9, 2025

• llama-chat : add phi 4 template (#11148)

  ngxson

  authored
  d9feae1

  Copy full SHA for d9feae1

• media : remove old img [no ci]

  ggerganov

  committed
  be0e950

  Copy full SHA for be0e950

• model: Add support for PhiMoE arch (#11003)

  Show description for f8feb4b
  

  phymbert

  and

  ThiloteE

  authored
  f8feb4b

  Copy full SHA for f8feb4b

• server : add tooltips to settings and themes btn (#11154)

  Show description for 8eceb88
  

  danbev

  and

  ngxson

  authored
  8eceb88

  Copy full SHA for 8eceb88

• doc: add cuda guide for fedora (#11135)

  Show description for 1204f97
  teihome

  authored
  1204f97

  Copy full SHA for 1204f97

Commits on Jan 10, 2025

• SYCL: Refactor ggml_sycl_compute_forward (#11121)

  Show description for c6860cc
  qnixsynapse

  authored
  c6860cc

  Copy full SHA for c6860cc

• llama: add support for QRWKV6 model architecture (#11001)

  Show description for ee7136c
  

  3 peopleauthored
  ee7136c

  Copy full SHA for ee7136c

• Vulkan: Fix float16 use on devices without float16 support + fix subgroup_size_control validation error (#11161)

  Show description for c3f9d25
  0cc4m

  authored
  c3f9d25

  Copy full SHA for c3f9d25

• convert : add --print-supported-models option (#11172)

  Show description for ff3fcab
  danbev

  authored
  ff3fcab

  Copy full SHA for ff3fcab

• examples : add README.md to tts example [no ci] (#11155)

  Show description for ba8a1f9
  danbev

  authored
  ba8a1f9

  Copy full SHA for ba8a1f9

Commits on Jan 11, 2025

• convert : sort print supported models [no ci] (#11179)

  Show description for 2739a71
  danbev

  authored
  2739a71

  Copy full SHA for 2739a71

• gguf-py: fixed local detection of gguf package (#11180)

  Show description for c05e8c9
  VJHack

  authored
  c05e8c9

  Copy full SHA for c05e8c9

Commits on Jan 12, 2025

• llama : add llama_vocab, functions -> methods, naming (#11110)

  Show description for afa8a9e
  

  ggerganov

  and

  slaren

  authored
  afa8a9e

  Copy full SHA for afa8a9e

• llama : remove notion of CLS token (#11064)

  Show description for 08f10f6
  ggerganov

  authored
  08f10f6

  Copy full SHA for 08f10f6

• llama : fix chat template gguf key (#11201)

  ngxson

  authored
  9a48399

  Copy full SHA for 9a48399

• Reset color before we exit (#11205)

  Show description for 924518e
  ericcurtin

  authored
  924518e

  Copy full SHA for 924518e

Commits on Jan 13, 2025

• ggml : do not define GGML_USE_CUDA when building with GGML_BACKEND_DL (#11211)

  Show description for 1244cdc
  rgerganov

  authored
  1244cdc

  Copy full SHA for 1244cdc

• llama : remove 'd' from bad special token log (#11212)

  Show description for 8f70fc3
  danbev

  authored
  8f70fc3

  Copy full SHA for 8f70fc3

• contrib : add naming guidelines (#11177)

  Show description for 7426a26
  ggerganov

  authored
  7426a26

  Copy full SHA for 7426a26

• common : support tag-based --hf-repo like on ollama (#11195)

  Show description for 00b4c3d
  ngxson

  authored
  00b4c3d

  Copy full SHA for 00b4c3d

• contrib : add naming guidelines (cont) (#11177)

  ggerganov

  authored
  ca001f6

  Copy full SHA for ca001f6

• server : (UI) Support for RTL text as models input or output (#11208)

  ebraminio

  authored
  437e05f

  Copy full SHA for 437e05f

• contrib : add naming guidelines (cont) (#11177)

  ggerganov

  committed
  a29f087

  Copy full SHA for a29f087

• cuda : CUDA Graph Compute Function Refactor (precursor for performance improvements) (#11042)

  Show description for 39509fb
  

  aendk

  and

  slaren

  authored
  39509fb

  Copy full SHA for 39509fb

• cli : auto activate conversation mode if chat template is available (#11214)

  Show description for 84a4481
  ngxson

  authored
  84a4481

  Copy full SHA for 84a4481

• server : (UI) Improve messages bubble shape in RTL (#11220)

  Show description for 504af20
  ebraminio

  authored
  504af20

  Copy full SHA for 504af20

Commits on Jan 14, 2025

• scripts : sync opencl

  ggerganov

  committed
  d00a80e

  Copy full SHA for d00a80e

• scripts : sync gguf

  ggerganov

  committed
  48e1ae0

  Copy full SHA for 48e1ae0

• scripts : sync gguf (cont)

  ggerganov

  committed
  a4f3f5d

  Copy full SHA for a4f3f5d

• sync : ggml

  ggerganov

  committed
  44d1e79

  Copy full SHA for 44d1e79

• Refactor test-chat-template.cpp (#11224)

  Show description for 091592d
  ochafik

  authored
  091592d

  Copy full SHA for 091592d

• server : Improve code snippets direction between RTL text (#11221)

  ebraminio

  authored
  c5bf0d1

  Copy full SHA for c5bf0d1

• vocab : add dummy tokens for "no_vocab" type (#11231)

  Show description for bbf3e55
  ggerganov

  authored
  bbf3e55

  Copy full SHA for bbf3e55

• ci : add -no-cnv for tests (#11238)

  ngxson

  authored
  b4d92a5

  Copy full SHA for b4d92a5

Commits on Jan 15, 2025

• SYCL: Add gated linear attention kernel (#11175)

  Show description for f446c2c
  qnixsynapse

  authored
  f446c2c

  Copy full SHA for f446c2c

• examples : add embd_to_audio to tts-outetts.py [no ci] (#11235)

  Show description for 0ccd7f3
  danbev

  authored
  0ccd7f3

  Copy full SHA for 0ccd7f3

• RoPE: fix back, CUDA support for back + noncont. (#11240)

  Show description for 432df2d
  JohannesGaessler

  authored
  432df2d

  Copy full SHA for 432df2d

• fix: ggml: fix vulkan-shaders-gen build (#10448)

  Show description for 1d85043
  sparkleholic

  authored
  1d85043

  Copy full SHA for 1d85043

• ci : use -no-cnv in gguf-split tests (#11254)

  Show description for f11cfdf
  ggerganov

  authored
  f11cfdf

  Copy full SHA for f11cfdf

• vulkan: scale caching for k quants + misc fixes (#11081)

  Show description for adc5dd9
  netrunnereve

  authored
  adc5dd9

  Copy full SHA for adc5dd9

Commits on Jan 16, 2025

• ggml: aarch64: implement SVE kernels for q4_K_q8_K vector dot (#11227)

  Show description for c67cc98
  

  fj-y-saito

  and

  ggerganov

  authored
  c67cc98

  Copy full SHA for c67cc98

• llama : add llama_model_load_from_splits (#11255)

  Show description for 681149c
  ngxson

  authored
  681149c

  Copy full SHA for 681149c

• CUDA: backwards pass for misc. ops, add tests (#11257)

  Show description for 9c8dcef
  JohannesGaessler

  authored
  9c8dcef

  Copy full SHA for 9c8dcef

• llama : add internlm3 support (#11233)

  Show description for 4dbc8b9
  RunningLeon

  authored
  4dbc8b9

  Copy full SHA for 4dbc8b9

• vulkan: optimize coopmat2 q2_k dequant function (#11130)

  jeffbolznv

  authored
  206bc53

  Copy full SHA for 206bc53

• vulkan: optimize coopmat2 q4_k/q5_k dequant functions. (#11206)

  Show description for 466300f
  jeffbolznv

  authored
  466300f

  Copy full SHA for 466300f

• vulkan: support copy from f32 to q4_0/q4_1/q5_0/q5_1/q8_0/iq4_nl (#11166)

  Show description for bd38dde
  jeffbolznv

  authored
  bd38dde

  Copy full SHA for bd38dde

Commits on Jan 17, 2025

• README : added kalavai to infrastructure list (#11216)

  musoles

  authored
  7a689c4

  Copy full SHA for 7a689c4

• llama : fix deprecation message: vocabable -> vocab (#11269)

  dwrensha

  authored
  960ec65

  Copy full SHA for 960ec65

• vocab : fix double-eos check (#11273)

  Show description for a133566
  ggerganov

  authored
  a133566

  Copy full SHA for a133566

• rpc : early register backend devices (#11262)

  Show description for 667d728
  rgerganov

  authored
  667d728

  Copy full SHA for 667d728

• llama.android: add field formatChat to control whether to parse special tokens when send message (#11270)

  codezjx

  authored
  3edfa7d

  Copy full SHA for 3edfa7d

Commits on Jan 18, 2025

• vulkan: fix coopmat2 flash attention for non-contiguous inputs (#11281)

  Show description for 44e18ef
  jeffbolznv

  authored
  44e18ef

  Copy full SHA for 44e18ef