Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit history

Commits on May 20, 2025

• mtmd-helper : bug fix to token batching in mtmd (#13650)

  Show description for b7a1746
  

  l3utterfly

  and

  ngxson

  authored
  b7a1746

  Copy full SHA for b7a1746

• kv-cache : add SWA support (#13194)

  Show description for e298d2f
  ggerganov

  authored
  e298d2f

  Copy full SHA for e298d2f

• sycl : Overcoming workaround for mmap() allocation on Windows (#13482)

  Show description for f7c9429
  s-Nick

  authored
  f7c9429

  Copy full SHA for f7c9429

Commits on May 19, 2025

• mtmd : add vision support for llama 4 (#13282)

  Show description for 92ecdcc
  ngxson

  authored
  92ecdcc

  Copy full SHA for 92ecdcc

Commits on May 17, 2025

• server : added --no-prefill-assistant flag (#13608)

  Show description for 6a2bc8b
  isaac-mcfadyen

  authored
  6a2bc8b

  Copy full SHA for 6a2bc8b

Commits on May 16, 2025

• server : do not return error out of context (with ctx shift disabled) (#13577)

  ngxson

  authored
  6aa892e

  Copy full SHA for 6aa892e

• webui : improve accessibility for visually impaired people (#13551)

  Show description for aea9f8b
  ngxson

  authored
  aea9f8b

  Copy full SHA for aea9f8b

Commits on May 15, 2025

• llama-bench : fix -ot with dl backends (#13563)

  slaren

  authored
  6c8b915

  Copy full SHA for 6c8b915

• webui : handle PDF input (as text or image) + convert pasted long content to file (#13562)

  Show description for 3cc1f1f
  ngxson

  authored
  3cc1f1f

  Copy full SHA for 3cc1f1f

• server : proper error handling for missing elements in messages array (OpenAI compatible backend) (#13540)

  pwilkin

  authored
  c753d7b

  Copy full SHA for c753d7b

• bench : handle decode errors (#13548)

  Show description for b283804
  ggerganov

  authored
  b283804

  Copy full SHA for b283804

• `server`: inject date_string in llama 3.x template + fix date for firefunction v2 (#12802)

  Show description for aa48e37
  

  3 peopleauthored
  aa48e37

  Copy full SHA for aa48e37

Commits on May 14, 2025

• `common`: add partial regex support (#12808)

  Show description for 3198405
  

  3 peopleauthored
  3198405

  Copy full SHA for 3198405

• server : passthrough the /models endpoint during loading (#13535)

  Show description for 0531744
  ggerganov

  authored
  0531744

  Copy full SHA for 0531744

• server : fix cache_tokens bug with no cache_prompt (#13533)

  ngxson

  authored
  360a9c9

  Copy full SHA for 360a9c9

• webui : use fflate for more deterministic gzip compress (#13525)

  Show description for bb1681f
  ngxson

  authored
  bb1681f

  Copy full SHA for bb1681f

• webui: Allow pasting file from clipboard (#13526)

  Show description for d486dd3
  

  luca020400

  and

  ngxson

  authored
  d486dd3

  Copy full SHA for d486dd3

Commits on May 13, 2025

• quantize : improve tensor-type pattern matching (#13033)

  EAddario

  authored
  e5c834f

  Copy full SHA for e5c834f

• clip : clip.h become private API (⚠️ breaking change) (#13510)

  ngxson

  authored
  71bdbdb

  Copy full SHA for 71bdbdb

• batched-bench : fix pp batch contents (#13492)

  ggerganov

  authored
  b89d605

  Copy full SHA for b89d605

• mtmd : remove libllava, remove clip-quantize-cli (⚠️ breaking change) (#13460)

  Show description for b472634
  ngxson

  authored
  b472634

  Copy full SHA for b472634

Commits on May 12, 2025

• llama-bench : add defrag-thold, check for invalid ranges (#13487)

  slaren

  authored
  cf0a43b

  Copy full SHA for cf0a43b

• clip : cap max image size 1024 for qwen vl model (#13478)

  ngxson

  authored
  de4c07f

  Copy full SHA for de4c07f

• server : allow content to be null in oaicompat_completion_params_parse (#13477)

  anudit

  authored
  91159ee

  Copy full SHA for 91159ee

• llama-bench : accept ranges for integer parameters (#13410)

  slaren

  authored
  22cdab3

  Copy full SHA for 22cdab3

Commits on May 11, 2025

• mtmd : Use RMS norm for InternVL 3 38B and 78B mmproj (#13459)

  city96

  authored
  c104023

  Copy full SHA for c104023

• tools : fix uninitialized llama_batch in server (#13436)

  Show description for 9a390c4
  

  aumfer

  and

  ngxson

  authored
  9a390c4

  Copy full SHA for 9a390c4

• Add `--no-op-offload` to improve `-ot` pp perf in MoE models like llama4 400B (#13386)

  hjc4869

  authored
  7f323a5

  Copy full SHA for 7f323a5

• mtmd : support InternVL 3 38B and 78B mmproj (#13443)

  Show description for 3eac209
  city96

  authored
  3eac209

  Copy full SHA for 3eac209

• mtmd : move helpers to dedicated file (#13442)

  Show description for a634d75
  ngxson

  authored
  a634d75

  Copy full SHA for a634d75

Commits on May 10, 2025

• mtmd : add hard limit on image resolution for qwen2vl / qwen2.5vl (#13434)

  Show description for 15e6125
  ngxson

  authored
  15e6125

  Copy full SHA for 15e6125

• server : update docs (#13432)

  ngxson

  authored
  3b24d26

  Copy full SHA for 3b24d26

• mtmd : support InternVL 2.5 and 3 (#13422)

  Show description for 053367d
  ngxson

  authored
  053367d

  Copy full SHA for 053367d

Commits on May 9, 2025

• server : vision support via libmtmd (#12898)

  Show description for 33eff40
  

  ngxson

  and

  ggerganov

  authored
  33eff40

  Copy full SHA for 33eff40

• llama : do not crash if there is no CPU backend (#13395)

  Show description for 27ebfca
  slaren

  authored
  27ebfca

  Copy full SHA for 27ebfca