Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit History

Commits on Dec 18, 2025

• llama: offload output layer to GPU first (#18148)

  JohannesGaessler

  authored
  57c1e05

  Copy full SHA for 57c1e05

• llama : Async DirectIO model loading on Linux (#18012)

  Show description for 4d4f4ca
  JTischbein

  authored
  4d4f4ca

  Copy full SHA for 4d4f4ca

Commits on Dec 17, 2025

• llama-fit-params: fix memory print (#18136)

  JohannesGaessler

  authored
  8dcc366

  Copy full SHA for 8dcc366

• common : restore grammar-based rejection sampling (#18137)

  Show description for 4301e27
  ggerganov

  authored
  4301e27

  Copy full SHA for 4301e27

• model: fix LFM2_MOE missing tensors (#18132)

  tdakhran

  authored
  982060f

  Copy full SHA for 982060f

Commits on Dec 16, 2025

• llama-fit-params: force disable mlock (#18103)

  JohannesGaessler

  authored
  d0794e8

  Copy full SHA for d0794e8

• llama-fit-params: lower ctx size for multi GPU (#18101)

  JohannesGaessler

  authored
  9dcac6c

  Copy full SHA for 9dcac6c

• llama-fit-params: fix underflow for dense models (#18095)

  JohannesGaessler

  authored
  0e49a7b

  Copy full SHA for 0e49a7b

• model: fix LFM2 missing tensors (#18105)

  ngxson

  authored
  ef83fb8

  Copy full SHA for ef83fb8

• llama: fix early stop in params_fit if ctx is set (#18070)

  JohannesGaessler

  authored
  ec98e20

  Copy full SHA for ec98e20

• arch: refactor LLM_TENSOR_NAMES (#18051)

  Show description for 7f2b2f3
  ngxson

  authored
  7f2b2f3

  Copy full SHA for 7f2b2f3

• Optimization: Qwen3 next autoregressive pass (#17996)

  Show description for a5251ca
  pwilkin

  authored
  a5251ca

  Copy full SHA for a5251ca

• model: support GLM4V vision encoder (#18042)

  Show description for 3d86c6c
  

  ngxson

  and

  ggerganov

  authored
  3d86c6c

  Copy full SHA for 3d86c6c

• llama: Include algorithm header needed for C++23 (#18078)

  cpeterso

  authored
  2aa45ef

  Copy full SHA for 2aa45ef

• graph : reuse SSM graphs (#16490)

  Show description for c560316
  ggerganov

  authored
  c560316

  Copy full SHA for c560316

• llama : add support for NVIDIA Nemotron 3 Nano (#18058)

  Show description for 2995341
  

  danbev

  and

  ggerganov

  authored
  2995341

  Copy full SHA for 2995341

Commits on Dec 15, 2025

• model : add KORMo model (#18032)

  Show description for 9d52f17
  HelloKS

  authored
  9d52f17

  Copy full SHA for 9d52f17

• kv-cache: Fix state restore fragmented cache (#17982)

  Show description for 4529c66
  

  ssweens

  and

  ggerganov

  authored
  4529c66

  Copy full SHA for 4529c66

• llama: automatically set parameters not set by the user in such a way that maximizes GPU utilization (#16653)

  Show description for b1f3a6e
  JohannesGaessler

  authored
  b1f3a6e

  Copy full SHA for b1f3a6e

Commits on Dec 14, 2025

• graph: add f_attn_temp_offset (#18025)

  ngxson

  authored
  0759b09

  Copy full SHA for 0759b09

• models : fix YaRN regression + consolidate logic (#18006)

  Show description for 609a2d0
  ggerganov

  authored
  609a2d0

  Copy full SHA for 609a2d0

Commits on Dec 13, 2025

• llama_context: synchronize before reallocating output buffer (#17974)

  jeffbolznv

  authored
  5266379

  Copy full SHA for 5266379

Commits on Dec 12, 2025

• models : fix the attn_factor for mistral3 graphs + improve consistency (#17945)

  Show description for 7bed317
  ggerganov

  authored
  7bed317

  Copy full SHA for 7bed317

Commits on Dec 11, 2025

• batch : fix sequence id ownership (#17915)

  Show description for d9f8f60
  ggerganov

  authored
  d9f8f60

  Copy full SHA for d9f8f60

Commits on Dec 10, 2025

• ggml : remove GGML_KQ_MASK_PAD constant (#17910)

  Show description for 4dff236
  ggerganov

  authored
  4dff236

  Copy full SHA for 4dff236

• model : Qwen3-Next-80B-A3B has 48 layers (#17898)

  Show description for b677721
  EZForever

  authored
  b677721

  Copy full SHA for b677721

Commits on Dec 9, 2025

• cmake: fix Mach-O current version number (#17877)

  Show description for 63908b6
  Rhys-T

  authored
  63908b6

  Copy full SHA for 63908b6

• model : nit, DeepSeek V1 MoE is 16B and GigaChat is 20B (#12652)

  Show description for 42b12b5
  CISC

  authored
  42b12b5

  Copy full SHA for 42b12b5

• llama : add token matching support to llama-grammar (#17816)

  Show description for e39502e
  aldehir

  authored
  e39502e

  Copy full SHA for e39502e

• model : support Rnj-1 (#17811)

  Show description for 1d2a1ab
  philip-essential

  authored
  1d2a1ab

  Copy full SHA for 1d2a1ab

Commits on Dec 8, 2025

• graph : use fill instead of scale_bias in grouped expert selection (#17867)

  Show description for c8554b6
  CISC

  authored
  c8554b6

  Copy full SHA for c8554b6

• Make graph_max_nodes vary by ubatch size (#17794)

  Show description for e4e9c43
  

  pwilkin

  and

  ggerganov

  authored
  e4e9c43

  Copy full SHA for e4e9c43

Commits on Dec 7, 2025

• model: add llama 4 scaling for mistral-large (deepseek arch) (#17744)

  ngxson

  authored
  4d37262

  Copy full SHA for 4d37262

Commits on Dec 6, 2025

• llama : remove quantization sanity check (#17788)

  Show description for 444f00b
  danbev

  authored
  444f00b

  Copy full SHA for 444f00b

Commits on Dec 5, 2025

• fix: prevent segfault in tokenizer on highly repetitive input (#17786)

  Show description for 1be9783
  ServeurpersoCom

  authored
  1be9783

  Copy full SHA for 1be9783