COMMITS
September 3, 2026
S
mla: use the absorbed path for multi-token steps (#2150)
Stuart Rowlands committed
A
Fix multi-turn crash on hybrid models with untrimmable caches (#2152)
Alazer Manakelew committed
A
Enable speculative decoding for Laguna (#2151)
Alazer Manakelew committed
A
Add Nemotron-H speculative decoding support (#2145)
Alazer Manakelew committed
September 2, 2026
P
Normalize tokenizer padding in BatchGenerator (#1951)
Prince Canuma committed
P
Keep quantized APC checkpoints packed (#2090)
Prince Canuma committed
P
Optimize exact Qwen verifier QMV unrolling (#2140)
Prince Canuma committed
A
Register the qwen3_vl_embedding auto-processor patch (#2146)
Alazer Manakelew committed
A
granite4_vision: align deepstack injection with chunked prefill via cache offset (#2134) (#2143)
Alazer Manakelew committed
Y
Add DeepSeek-V4 Flash Vision EXP support (#2136)
Yoan Ballesteros committed
P
Merge pull request #2142 from Micha0827/fix/qwen3-vl-video-timestamps
Prince Canuma committed
M
qwen3_vl: render per-group video timestamps like the reference processor
Michael Gross committed
September 1, 2026
A
Add DSpark drafter for Gemma 4 (#2137)
Alazer Manakelew committed
P
Merge pull request #2133 from Blaizzy/pc/bump-mlx-0.32.2
Prince Canuma committed
P
chore: require MLX 0.32.2
Prince Canuma committed
L
Merge pull request #2132 from lucasnewman/support-multiple-tool-calls
Lucas Newman committed
L
Support multiple tool call invocations at once.
Lucas Newman committed
L
Merge pull request #2131 from lucasnewman/oai-privacy-filter
Lucas Newman committed
L
Add OpenAI Privacy Filter model support.
Lucas Newman committed
A
sam3: fix pred_boxes inflation and offset (RoPE scale, RPB grid, ref_point_head) (#2128)
Alazer Manakelew committed
A
Strip Muse Glimmer channel scaffolding from generated output (#2121) (#2129)
Alazer Manakelew committed
L
Merge pull request #2104 from eklipse2k8/feat/video-depth-anything
Lucas Newman committed
M
Merge branch 'main' into feat/video-depth-anything
Matt Jarjoura committed
A
fix(florence2): load legacy large-ft checkpoints on transformers>=5.14 (#2123)
Alazer Manakelew committed
A
Add OmniParser v2 YOLO11 support (#2122)
Alazer Manakelew committed
L
Merge pull request #1822 from YuhuaBillChen/pr1-tq-mtp
Lucas Newman committed
A
Merge branch 'main' into pr1-tq-mtp
Alazer Manakelew committed
August 31, 2026
M
Add Video Depth Anything (video_depth_anything)
Matt Jarjoura committed
A
Add GLM-5.3 MTP support (#2079)
Alazer Manakelew committed
A
GLM-5.3 - 2026-08-28 (#2078)
Alazer Manakelew committed