Commits

Breadcrumbs

History for

onmain

User selector

All users

All time

Commit History

Commits on Jun 21, 2026

• feat(speculative): Improve ngram-map draft selection and accept feedback

  Show description for acc3fe2
  JamePeng

  committed
  acc3fe2

  Copy full SHA for acc3fe2

• Update Submodule vendor/llama.cpp bfa3219..bddfd2b

  Show description for a27b100
  JamePeng

  committed
  a27b100

  Copy full SHA for a27b100

• Sync upstream: common/json-schema-to-grammar : align spacing rules with parsers (#24835)

  Show description for f566d6b
  JamePeng

  committed
  f566d6b

  Copy full SHA for f566d6b

Commits on Jun 19, 2026

• Update Submodule vendor/llama.cpp f3e1828..db52540

  Show description for b4f2d96
  JamePeng

  committed
  b4f2d96

  Copy full SHA for b4f2d96

Commits on Jun 16, 2026

• Update mtmd API 20260616

  Show description for 6a48d58
  JamePeng

  committed
  6a48d58

  Copy full SHA for 6a48d58

Commits on Jun 14, 2026

• feat(mtmd): broaden multimodal media extraction

  Show description for 284df51
  JamePeng

  committed
  284df51

  Copy full SHA for 284df51

• feat(mtmd): enhance generic chat template support

  Show description for c974531
  JamePeng

  committed
  c974531

  Copy full SHA for c974531

• fix(model): handle missing chat templates

  Show description for d84b0c2
  JamePeng

  committed
  d84b0c2

  Copy full SHA for d84b0c2

• refactor(mtmd): move multimodal handlers to separate module `llama_multimodal`

  Show description for 1965d5f
  JamePeng

  committed
  1965d5f

  Copy full SHA for 1965d5f

• Merge branch 'main' into mtmd

  alcoftTAO

  authored
  5757856

  Copy full SHA for 5757856

• Change 'clip_model_path' to 'mmproj_path'. Implemented 'chat_template_override'. Only the chat template is passed from llama to the chat handler; not the entire model's metadata.

  alcoftTAO

  committed
  d8ee3ee

  Copy full SHA for d8ee3ee

Commits on Jun 13, 2026

• Update(MTMDChatHandler): add chunk type helpers

  Show description for cb299e6
  JamePeng

  committed
  cb299e6

  Copy full SHA for cb299e6

• Update(mtmd): Append mtmd batching API

  Show description for 971ee38
  JamePeng

  committed
  971ee38

  Copy full SHA for 971ee38

Commits on Jun 9, 2026

• Update Submodule vendor/llama.cpp 7d2b45b..d6d0ce8

  Show description for e4dcac1
  JamePeng

  committed
  e4dcac1

  Copy full SHA for e4dcac1

Commits on Jun 8, 2026

• feat(mtmd): add video input support to MTMDChatHandler

  Show description for 10b4add
  JamePeng

  committed
  10b4add

  Copy full SHA for 10b4add

• Update mtmd API 20260609

  Show description for 55e855b
  JamePeng

  committed
  55e855b

  Copy full SHA for 55e855b

Commits on Jun 7, 2026

• feat(_ctypes_extensions): improve error diagnostics for shared library loading

  Show description for 7a8272e
  JamePeng

  committed
  7a8272e

  Copy full SHA for 7a8272e

• Update Submodule vendor/llama.cpp f71af35..f0156d1

  Show description for b1ad445
  JamePeng

  committed
  b1ad445

  Copy full SHA for b1ad445

• Bump version to 0.3.40-Milestone

  Show description for 12861b9
  JamePeng

  committed
  12861b9

  Copy full SHA for 12861b9

Commits on Jun 6, 2026

• feat(chat-format): Update google/gemma-4 chat template jinja

  Show description for 5929d86
  JamePeng

  committed
  5929d86

  Copy full SHA for 5929d86

• fix(mtmd): memory_can_shift() logic bug

  Show description for e001886
  JamePeng

  committed
  e001886

  Copy full SHA for e001886

• Update Submodule vendor/llama.cpp c4a278d..6b80c74

  Show description for fb65ed7
  JamePeng

  committed
  fb65ed7

  Copy full SHA for fb65ed7

• feat: pass reasoning budget params through Llama APIs

  Show description for 1b472b3
  JamePeng

  committed
  1b472b3

  Copy full SHA for 1b472b3

• feat(internals): add verbose logging to ReasoningBudgetSampler

  Show description for 82a0266
  JamePeng

  committed
  82a0266

  Copy full SHA for 82a0266

Commits on Jun 5, 2026

• feat(internals): Add `ReasoningBudgetSampler` support

  Show description for be123f1
  JamePeng

  committed
  be123f1

  Copy full SHA for be123f1

Commits on Jun 2, 2026

• Update Submodule vendor/llama.cpp 27d9ed8..60130d1

  Show description for acf8963
  JamePeng

  committed
  acf8963

  Copy full SHA for acf8963

Commits on Jun 1, 2026

• feat(llama): enhance chat template initialization with full special tokens

  Show description for bbede19
  JamePeng

  committed
  bbede19

  Copy full SHA for bbede19

• feat(chat-format): improve Jinja2ChatFormatter HF compatibility

  Show description for a7db23a
  JamePeng

  committed
  a7db23a

  Copy full SHA for a7db23a

• Update Submodule vendor/llama.cpp d4c8e2c..27d9ed8

  Show description for 9bb06da
  JamePeng

  committed
  9bb06da

  Copy full SHA for 9bb06da

• feat(sampling): add reasoning budget configurations

  Show description for e174c10
  JamePeng

  committed
  e174c10

  Copy full SHA for e174c10

Commits on May 31, 2026

• feat: add `ReasoningBudgetState` enum and `TokenMatcher` helper class to _internals.py

  Show description for ca81fd4
  JamePeng

  committed
  ca81fd4

  Copy full SHA for ca81fd4

• refactor(llama_cpp): wrap llama constants into enum.IntEnum

  Show description for ea0907d
  JamePeng

  committed
  ea0907d

  Copy full SHA for ea0907d

Commits on May 29, 2026

• feat(mtmd): improve fallback chat template for multimodal models

  Show description for e7976f4
  JamePeng

  committed
  e7976f4

  Copy full SHA for e7976f4

• feat(chat_handler): update multimodal handlers for Qwen2.5-VL, Qwen3-VL, and PaddleOCR

  Show description for 6c9e7bf
  JamePeng

  committed
  6c9e7bf

  Copy full SHA for 6c9e7bf

Commits on May 27, 2026

• Added support when using the keyword 'audio' instead of 'audio_url'.

  alcoftTAO

  committed
  4794c8c

  Copy full SHA for 4794c8c