Commits

Branch selector

 main

User selector

All users

All time

Commit History

Commits on Mar 22, 2026

• fix(types): correct `llama_set_adapters_lora` LoRA adapter ctypes signature and use pointer for scales

  Show description for e42cf39
  JamePeng

  committed
  e42cf39

  Copy full SHA for e42cf39

• refactor: remove legacy static LoRA initialization

  Show description for 87157f3
  JamePeng

  committed
  87157f3

  Copy full SHA for 87157f3

• Update Submodule vendor/llama.cpp fb78ad2..3306dba

  JamePeng

  committed
  801755e

  Copy full SHA for 801755e

Commits on Mar 20, 2026

• chore: enhance hybrid cache logging and document M-RoPE token usage

  Show description for d2318a4
  JamePeng

  committed
  d2318a4

  Copy full SHA for d2318a4

• Update Submodule vendor/llama.cpp 312cf03..fb78ad2

  JamePeng

  committed
  bdc2d7c

  Copy full SHA for bdc2d7c

Commits on Mar 18, 2026

• Update Submodule vendor/llama.cpp d23355a..312cf03

  Show description for ed5e212
  JamePeng

  committed
  ed5e212

  Copy full SHA for ed5e212

Commits on Mar 15, 2026

• feat(core): add verbose debug logging to longest_token_prefix fast paths

  Show description for 90d46dc
  JamePeng

  committed
  90d46dc

  Copy full SHA for 90d46dc

• Update MIT license copyright to collective authorship (2023-2026)

  Show description for 35957bd
  JamePeng

  committed
  35957bd

  Copy full SHA for 35957bd

• Update CMakeLists.txt

  JamePeng

  committed
  9c29c27

  Copy full SHA for 9c29c27

• Bump version to 0.3.33

  Show description for 6bbc8d2
  JamePeng

  committed
  6bbc8d2

  Copy full SHA for 6bbc8d2

Commits on Mar 14, 2026

• Update Submodule vendor/llama.cpp 463b6a9..d23355a

  JamePeng

  committed
  b0f00a9

  Copy full SHA for b0f00a9

• perf(mtmd): optimize media_id masking with bitwise AND

  Show description for bd0b069
  JamePeng

  committed
  bd0b069

  Copy full SHA for bd0b069

• fix(mtmd): prevent multimodal image freeze by injecting deterministic media IDs

  Show description for dd111eb
  JamePeng

  committed
  dd111eb

  Copy full SHA for dd111eb

• Update Submodule vendor/llama.cpp 8f974d2..463b6a9

  JamePeng

  committed
  ec2cbea

  Copy full SHA for ec2cbea

Commits on Mar 13, 2026

• Update mtmd_cpp API 20260313

  JamePeng

  committed
  afd06ea

  Copy full SHA for afd06ea

• Update Submodule vendor/llama.cpp 557fe2d..8f974d2

  JamePeng

  committed
  a6db3f2

  Copy full SHA for a6db3f2

• fix(Llama.generate): add explicit fallback context reset and expand `Llama.generate` docstrings

  Show description for 6f81466
  JamePeng

  committed
  6f81466

  Copy full SHA for 6f81466

Commits on Mar 12, 2026

• docs(README.md): add sampling parameter guide and strategic project tips to README

  Show description for d422e82
  JamePeng

  committed
  d422e82

  Copy full SHA for d422e82

• fix(sampling): prevent memory drift and hallucinations in logits view

  Show description for aa38850
  JamePeng

  committed
  aa38850

  Copy full SHA for aa38850

• Update Submodule vendor/llama.cpp fdb1764..557fe2d

  JamePeng

  committed
  380c643

  Copy full SHA for 380c643

Commits on Mar 11, 2026

• feat(_ggml): extend ctypes bindings with more ggml constants, enums, and structs

  JamePeng

  committed
  2bb5cd9

  Copy full SHA for 2bb5cd9

• Update llama.cpp API 20260312

  JamePeng

  committed
  c420a2a

  Copy full SHA for c420a2a

• Update Submodule vendor/llama.cpp 182acfe..fdb1764

  JamePeng

  committed
  21c4c87

  Copy full SHA for 21c4c87

• fix(chat_handler): fix tools and function calling in MTMDChatHandler.

  JamePeng

  committed
  8641667

  Copy full SHA for 8641667

• Update Submodule vendor/llama.cpp 10e5b14..182acfe

  JamePeng

  committed
  b827932

  Copy full SHA for b827932

Commits on Mar 10, 2026

• perf(cache): optimize LlamaDiskCache I/O and fix LRU behavior

  Show description for a36e8d2
  JamePeng

  committed
  a36e8d2

  Copy full SHA for a36e8d2

• Sync examples : fix empty items in json_schema_to_grammar.py (#19968)

  JamePeng

  committed
  8e8f36f

  Copy full SHA for 8e8f36f

• Update Submodule vendor/llama.cpp ec947d2..10e5b14

  JamePeng

  committed
  054f04e

  Copy full SHA for 054f04e

• perf(cache): upgrade `LlamaRAMCache` to O(1) eviction and set `LlamaTrieCache` as default

  Show description for 5c8e056
  JamePeng

  committed
  5c8e056

  Copy full SHA for 5c8e056

• Update Submodule vendor/llama.cpp 59db9a3..ec947d2

  JamePeng

  committed
  4bec165

  Copy full SHA for 4bec165

Commits on Mar 9, 2026

• fix(core): disable swa_full for non-SWA models (sync llama.cpp upstream #20291)

  Show description for 9acc507
  JamePeng

  committed
  9acc507

  Copy full SHA for 9acc507

• Update llama.cpp API 20260310

  JamePeng

  committed
  955ac33

  Copy full SHA for 955ac33

• fix(chat_format): fix namespace and variable shadowing of llama modules

  Show description for 7efcb2b
  JamePeng

  committed
  7efcb2b

  Copy full SHA for 7efcb2b

• fix(cache): fix namespace shadowing to prevent AttributeError

  Show description for 5e285fe
  JamePeng

  committed
  5e285fe

  Copy full SHA for 5e285fe

• Update Submodule vendor/llama.cpp 35bee03..59db9a3

  JamePeng

  committed
  9cc2efb

  Copy full SHA for 9cc2efb