Skip to content

Pull requests: Blaizzy/mlx-vlm

New pull request New
67 Open 1,165 Closed
67 Open 1,165 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

perf(qwen3.8): adapt MTP draft depth
#2046 opened Aug 27, 2026 by Thump604 Contributor • Draft
1
feat(qwen3.8): add external PLE storage
#2045 opened Aug 27, 2026 by Thump604 Contributor • Draft
2
MTP support for GLM-5.3-Flash
#2044 opened Aug 27, 2026 by Lazarus-931 Collaborator • Draft
Drop unused DeepSeek-V4 rotating-cache value graphs
#2043 opened Aug 27, 2026 by byoan Contributor Loading…
1
1
Fix Qwen4-Exp QSA cache for continuous batching
#2039 opened Aug 26, 2026 by jonathan308 Loading…
5
fix(cache): trim ChunkedKVCache on valid length, not buffer width (#2031)
#2036 opened Aug 26, 2026 by Anai-Guo Contributor Loading…
1
2
Fix positioned top-k sampling for speculative generation
#2029 opened Aug 26, 2026 by tjansn Loading…
fix(server): stop dropping streamed content after a tool call ends
#2027 opened Aug 26, 2026 by Anai-Guo Contributor Loading…
2
fix(sam3): let --threshold fall through to the documented per-task defaults
#2025 opened Aug 26, 2026 by Anai-Guo Contributor Loading…
fix(rfdetr): honor --show-boxes instead of ignoring it
#2024 opened Aug 26, 2026 by Anai-Guo Contributor Loading…
3
Support DeepSeek V4 speculative chunked prefill
#2023 opened Aug 26, 2026 by byoan Contributor Loading…
1
3
Fix Gemma 4 Unified HF vision weight mapping
#2018 opened Aug 24, 2026 by Supersynergy Loading…
2
Report speculative usage from direct generation
#1997 opened Aug 21, 2026 by byoan Contributor Loading…
1
@Blaizzy
Apply max_kv_size to models that build their own cache
#1995 opened Aug 21, 2026 by Lazarus-931 Collaborator Loading…
6
reformatting docs: move the README into a docs/ site
#1964 opened Aug 19, 2026 by Lazarus-931 Collaborator • Draft
Redesign APC for dense and hybrid model caches
#1960 opened Aug 19, 2026 by Blaizzy Owner Loading…
14
Fix speculative decoding against a quantized KV cache
#1956 opened Aug 19, 2026 by Codcore Loading…
@Blaizzy
1
Normalize tokenizer padding in BatchGenerator
#1951 opened Aug 18, 2026 by Blaizzy Owner • Draft
1
Stop Kimi-VL generation at turn boundary tokens
#1944 opened Aug 18, 2026 by leepokai Loading…
1
Reformat trainer for all modalities
#1939 opened Aug 17, 2026 by Goekdeniz-Guelmez Contributor • Draft
2
Fix Qwen speculative decoding with quantized batch cache
#1938 opened Aug 17, 2026 by Ptico Loading…
3
fix(server): remove the thought prefix by prefix, not by character set
#1934 opened Aug 17, 2026 by Anai-Guo Contributor Loading…
3
Key vision feature cache on content, not image path
#1933 opened Aug 17, 2026 by evnchn Loading…
5
feat(server): add conservative DFlash APC prefix reuse
#1923 opened Aug 16, 2026 by pierre427 • Draft
1
Add native DSpark speculative decoding
#1914 opened Aug 15, 2026 by Blaizzy Owner • Draft
ProTip! Add no:assignee to see everything that’s not assigned.