Skip to content

Pull requests: Blaizzy/mlx-vlm

New pull request New
68 Open 1,358 Closed
68 Open 1,358 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

feat(apc): reuse Qwen image prefixes when appending images
#2309 opened Sep 19, 2026 by kiarina Contributor Loading…
Add Mistral Large 3 (675B VLM) support
#2308 opened Sep 19, 2026 by Lazarus-931 Collaborator Loading…
1
Fix Qwen3-VL position embeddings with a quantized pos_embed
#2304 opened Sep 18, 2026 by rcfa Loading…
fix: keep converted sidecars writable and record the cast dtype
#2298 opened Sep 18, 2026 by ykhrustalev Contributor Loading…
Respect generation config sampling defaults enhancement New feature or request
#2289 opened Sep 17, 2026 by JakeATX Contributor Loading…
1
@Blaizzy
1
fix(turboquant): gate single-pass MSE decode at dim<=256 and fallback on threadgroup limits
#2288 opened Sep 17, 2026 by casualuser Loading…
1
feat: SAM 3D Objects for MLX
#2282 opened Sep 16, 2026 by eklipse2k8 Contributor Loading…
7
Expose make_batch_cache for schedulers outside mlx-vlm
#2279 opened Sep 16, 2026 by LXD-8 Loading…
@Blaizzy
6
Fix cancellation during multi-row prefill
#2273 opened Sep 16, 2026 by Blaizzy Owner • Draft
fix(speculative): allow chunked GLM MTP prefill
#2268 opened Sep 16, 2026 by raullenchai Contributor Loading…
2
feat(speculative): support GLM-5.3 DFlash2 targets
#2267 opened Sep 16, 2026 by raullenchai Contributor Loading…
3
Move a finished conversation's prefix cache to disk
#2261 opened Sep 15, 2026 by Lazarus-931 Collaborator Loading…
Fix Chat and Responses conversation parity
#2260 opened Sep 15, 2026 by Blaizzy Owner Loading…
glm5_next: opt-in compiled cached KDA decode step (bit-exact, +2% decode)
#2255 opened Sep 14, 2026 by freddyhaddad Loading…
1
Video Depth Anything: add opt-in fused Metal preprocessing
#2248 opened Sep 14, 2026 by aprylewu Loading…
fix(glm5-next): remap quantized lm head parameters
#2231 opened Sep 13, 2026 by raullenchai Contributor Loading…
4
Support thinking budgets during MTP speculative decoding
#2230 opened Sep 13, 2026 by YuhuaBillChen Contributor Loading…
Preserve message metadata through dictionary formatting
#2223 opened Sep 12, 2026 by bstpierr • Draft
1
fix(turboquant): drop shadowed duplicate BatchTurboQuantKVCache.zero_row_tail
#2222 opened Sep 11, 2026 by Anai-Guo Contributor Loading…
2
DeepSeek V4.1 vision and language model support
#2214 opened Sep 10, 2026 by Lazarus-931 Collaborator Loading…
6
Preserve image ownership across Chat Completions turns
#2212 opened Sep 10, 2026 by bstpierr Loading…
1
Rebuild speculative decoding around request caches and shared prefill
#2206 opened Sep 9, 2026 by Blaizzy Owner Loading…
11
Forward explicit reasoning-history policy to chat templates
#2200 opened Sep 9, 2026 by bstpierr Loading…
Preserve Qwen assistant reasoning through prompt normalization
#2199 opened Sep 9, 2026 by bstpierr • Draft
1
Fix qwen3_5 ragged speculative rollback on BatchQSAKVCache
#2197 opened Sep 9, 2026 by gaineyllc Loading…
2
ProTip! Adding no:label will show everything without a label.