Skip to content

Pull requests: ml-explore/mlx-lm

New pull request New
249 Open 689 Closed
249 Open 689 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

Add KV cache quantization to mlx_lm.server (#1043)
#1476 opened Jul 5, 2026 by katlun-lgtm Loading…
2 of 3 tasks
1
1
Fix NewlineTokenizer registration with Transformers v5
#1474 opened Jul 5, 2026 by kime541200 Loading…
fix(vl): strip model.visual.* in qwen2_vl/qwen3_vl/qwen3_vl_moe sanitize
#1473 opened Jul 4, 2026 by Jonathangadeaharder Loading…
fix: pin transformers<5.13.0 to avoid AutoTokenizer.register breakage
#1471 opened Jul 4, 2026 by rggammon Loading…
Fix MTPHead: correct hidden-state/embedding concat order and use pre-norm hidden
#1469 opened Jul 4, 2026 by h9q2cyxvgm-ui Loading…
Attach Qwen3.6 MTP head for speculative decoding + fix cache-rewind for hybrid recurrent models
#1468 opened Jul 4, 2026 by h9q2cyxvgm-ui Loading…
1
1
Fix broadcast crash in quantized SDPA with GQA + batched padding mask (batch >= 2)
#1467 opened Jul 4, 2026 by pinglin Loading…
Add quant2 / quant2_128 mixed-bit quant recipes
#1466 opened Jul 4, 2026 by dahai80 Loading…
Fix NewlineTokenizer registration for transformers >= 5.13
#1465 opened Jul 4, 2026 by chandukona Loading…
1
Add LongCat-2.0
#1464 opened Jul 4, 2026 by kernelpool Contributor Loading…
GLM-5.2: full/shared indexer typing for glm_moe_dsa (DSA schedule + interleaved indexer rope)
#1463 opened Jul 4, 2026 by machiabeli • Draft
Fix import crash with transformers >= 5.13
#1460 opened Jul 3, 2026 by Lazarus-931 Loading…
1
Fix AutoTokenizer.register() for transformers 5.13.0+ compatibility
#1459 opened Jul 3, 2026 by jonpspri Loading…
1
qwen3_5: load in-checkpoint MTP head + speculative rollback for hybrid (GDN) caches
#1456 opened Jul 3, 2026 by pierre427 Loading…
5
fix(mlx_lm.server): fail fast when --draft-model set with non-trimmable cache
#1455 opened Jul 2, 2026 by tejkas Loading…
1
DeepSeek-V3.2/GLM DSA: fix silent >128k top-k corruption + sparse-gather prefill
#1454 opened Jul 2, 2026 by aidiffuser Loading…
Fix DSA indexer LoRA-training crash: stop gradients through sparse-attention top-k indices
#1452 opened Jul 2, 2026 by trevorgordon981 Loading…
1
Fix Mistral tool parser dropping parallel/multiple tool calls
#1448 opened Jul 2, 2026 by DavidObando Loading…
Fix dropped tool calls for models with empty tool_call_end (Mistral/Devstral)
#1447 opened Jul 1, 2026 by DavidObando Loading…
1
Fix frozen PRNG in categorical_sampling under repeated sampling
#1444 opened Jun 30, 2026 by utkarshtiwari-24 Loading…
1
2
Fix qwen3.5-MoE garbage output: don't double-shift RMSNorm on MTP-retaining checkpoints
#1442 opened Jun 29, 2026 by embwl0x Loading…
Feature/layer streaming
#1440 opened Jun 28, 2026 by SashimiSaketoro Loading…
Make RotatingKVCache trimmable so prefix cache reuse works for sliding-window models
#1437 opened Jun 26, 2026 by amirarsalan90 Loading…
Fix: pythonic tool parser not auto-detected for LFM2.5 models
#1436 opened Jun 25, 2026 by grumdahl Loading…
fix: use FiscalNote/billsum HF dataset path in test_datsets
#1434 opened Jun 25, 2026 by ttxs69 Loading…
ProTip! Exclude everything labeled bug with -label:bug.