Skip to content

Pull requests: Blaizzy/mlx-vlm

New pull request New
49 Open 512 Closed
49 Open 512 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

Fix DFlash speculative decoding: GPU hang, performance, and upstream alignment
#1053 opened Apr 23, 2026 by Blaizzy Owner Loading…
4 of 6 tasks
Fix Gemma 4 LoRA training: vision backward NaN + audio_tower freeze leak
#1052 opened Apr 23, 2026 by john-rocky Loading…
Thread-local generation stream (port mlx-lm#1090)
#1050 opened Apr 22, 2026 by Blaizzy Owner Loading…
4 tasks
Add server json_schema response_format support
#1047 opened Apr 22, 2026 by avbiswas Loading…
1
30
Fix Kimi VL concurrent Metal crash and mixed-batch text degradation
#1039 opened Apr 20, 2026 by Blaizzy Owner Loading…
4 tasks
Add vision feature caching to all models
#1028 opened Apr 16, 2026 by Blaizzy Owner Loading…
6 tasks done
Expose presence_penalty, frequency_penalty, and per-penalty context_size on the server API
#1023 opened Apr 14, 2026 by esaruoho Loading…
1
refactor: improve model loading and resource handling in utils.py
#1019 opened Apr 13, 2026 by SyedaAnshrahGillani Loading…
fix: propagate the verbose to the Prefill tqdm
#1015 opened Apr 12, 2026 by PeterStaar-IBM Loading…
1
server: indicate finish reason properly when model made a tool call.
#1014 opened Apr 12, 2026 by viktike Contributor Loading…
perf: close 5.5% decode gap vs mlx_lm.server on streaming chat endpoint
#1012 opened Apr 11, 2026 by chilang Loading…
fix: use OpenAI chat-completion field names in /chat/completions usage
#1009 opened Apr 10, 2026 by chilang Loading…
fix: replace NaN from all-masked SDPA padding rows in Gemma 4 vision
#1006 opened Apr 10, 2026 by fabiopili Loading…
4 tasks done
feat: OpenAI Responses API with structured tool calling and multi-turn support
#996 opened Apr 9, 2026 by eloe Loading…
feat: add JSON mode via response_format parameter
#993 opened Apr 9, 2026 by eloe Loading…
feat: enforce tool_choice parameter in chat/completions
#992 opened Apr 9, 2026 by eloe Loading…
feat: add stop sequences support for both endpoints
#991 opened Apr 9, 2026 by eloe Loading…
Fix sft_trainer crash when adapter_file is None (#908)
#987 opened Apr 9, 2026 by H-A-Khan Loading…
1
Fix LoRA scaling: divide alpha by rank (#845)
#986 opened Apr 9, 2026 by H-A-Khan Loading…
1
Add TriAttention KV cache compression
#985 opened Apr 9, 2026 by Blaizzy Owner Loading…
3 of 4 tasks
Fix for Gemma4 audio task
#980 opened Apr 8, 2026 by lifeiteng Loading…
fix(trainer): pass images to prepare_inputs for Gemma, Qwen, and SmolVLM
#979 opened Apr 8, 2026 by ukint-vs Loading…
fix(trainer): flatten input_ids before measuring length in batch padding
#977 opened Apr 8, 2026 by ukint-vs Loading…
3 tasks done
Fix rotation-basis mismatch in TurboQuant L=1 fast quantize
#973 opened Apr 7, 2026 by nnorris7 Loading…
3 of 4 tasks
fix(generate): auto-apply chat template for vision/audio prompts
#972 opened Apr 7, 2026 by ukint-vs Loading…
ProTip! Adding no:label will show everything without a label.