electron-rare/mlx-lm
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Commit Graph

  • 54449a5056 feat(tuner): support shared 2D LoRA adapters on SwitchLinear feat/gpt-oss-bf16-and-moe-lora electron-rare 2026-06-02 00:10:25 +02:00
  • 88d34b3564 fix(gpt-oss): support unsloth/HF BF16 weight layout in sanitize electron-rare 2026-06-02 00:10:13 +02:00
  • 83c408e1e6 log only on rank 0 ui Anastasiia Filippova 2026-05-26 20:57:22 +02:00
  • 5b56e28af8 only light theme for now Anastasiia Filippova 2026-05-26 16:15:00 +02:00
  • c12cb31faa fix distibuted probing for the terminal theme Anastasiia Filippova 2026-05-26 15:55:57 +02:00
  • be13b3f735 UI for lora and chat Anastasiia Filippova 2026-05-26 13:49:35 +02:00
  • a716f040b7 Fix tokenizer test failure fix-tokenizer-test Cheng 2026-05-19 10:02:40 +09:00
  • ecf87bcf21 added tqdm to save_model save-tqdm Anastasiia Filippova 2026-05-14 20:00:29 +02:00
  • 92381c4eeb log only on rank 0 log-only-on-0 Anastasiia Filippova 2026-05-13 13:49:39 +02:00
  • df1d3f3c9a Fix Gemma 4 sanitize() not stripping KV projections for shared layers (#1240) main Arun Raj 2026-05-05 03:56:18 +05:30
  • 95238e5d34 sharded gemma 4 sharded_gemma4 Anastasiia Filippova 2026-04-23 15:53:57 +02:00
  • ed1fca4cef Thread local generation stream (#1090) v0.31.3 Angelos Katharopoulos 2026-04-22 00:34:09 -07:00
  • 4f5cbd2a4f Fix Gemma 4 KV-shared layers creating unused projections (#1158) glyphVault 2026-04-21 16:44:13 -07:00
  • 3cd9a52df2 Fix ArraysCache extend (#1177) Angelos Katharopoulos 2026-04-21 16:41:49 -07:00
  • 2f1ab85aec Fix Mistral empty tool_call_end flipping state machine to normal (#1151) Eyüp Can Akman 2026-04-21 11:36:56 +03:00
  • f3bb10c141 Fix Gemma4 tool parser: support hyphenated names and braces in strings (#1150) AkashKhamkar 2026-04-21 13:45:58 +05:30
  • e1c24b3237 fix: handle NoneType check for think tokens in TokenizerWrapper (#1167) Sherry Lo 2026-04-21 16:13:15 +08:00
  • f39cb8e934 Fix dwq: check for actual safetensors in target_dir (#1173) Luis Molina 2026-04-21 09:41:53 +02:00
  • a9856b485d Fix batch dimension mismatch in ArraysCache extend() (#1169) TechToboggan 2026-04-21 02:42:03 -04:00
  • e92138cb01 Apertus tie_word_embeddings fix (#1143) Andrei Panferov 2026-04-20 23:47:40 +02:00
  • a401730941 Fix missing tree_reduce import in models/cache.py (#1165) Siiea-ai 2026-04-20 13:31:58 -05:00
  • 6d114686e5 Fix MiniMax M2 parallel tool calling (#1171) Tarjei Mandt 2026-04-20 13:57:20 -04:00
  • aa4f880fb3 Fix parallel tool call handling in server (#1170) Tarjei Mandt 2026-04-20 02:58:25 -04:00
  • 62f38aeb51 Fix batch dimension mismatch in BatchKVCache and BatchRotatingKVCache extend() (#1141) razorback16 2026-04-14 17:21:31 -07:00
  • d9c63fff67 Bump the patch version (#1124) Angelos Katharopoulos 2026-04-08 02:04:27 -07:00
  • dcbf6e33d1 Align batch logits processor token contract (#1115) v0.31.2 Neil Mehta 2026-04-06 21:07:35 -04:00
  • f26fddfd3b Gemma4 final fixes and multi-token think/tool start/end (#1114) Angelos Katharopoulos 2026-04-06 17:40:48 -07:00
  • f56d99712c Fix output corruption in speculative decoding (#1109) Tarjei Mandt 2026-04-07 09:00:38 +10:00
  • c65c27b450 Fix Gemma 4 quantized per-layer projection loading (#1112) spicyneuron 2026-04-07 04:26:07 +08:00
  • 3257c3df17 Add Gemma 4 tool call parser (#1105) Nic Davidson 2026-04-04 18:21:45 -06:00
  • d4eb136d44 Bring back max-kv-size to the batch generator (#1106) Angelos Katharopoulos 2026-04-04 17:12:43 -07:00
  • 4469ad4647 Add gemma 4 (#1093) Prince Canuma 2026-04-04 13:47:03 +02:00
  • f79dba7832 perf: use max instead of argsort in apply_min_p sampling (#1083) Matteo Celani 2026-04-02 01:26:59 +02:00
  • 3f9d179fd1 Batch generation refactoring and various fixes (#1072) Angelos Katharopoulos 2026-04-01 15:07:50 -07:00
  • 9dc023beed Fix PromptTrie.pop_prefixes() off-by-one when pruning immediate prefixes (#1078) Lik Xun Yuan (Lx) 2026-04-02 02:51:19 +08:00
  • 9dcefa5272 fix: break shared-buffer memory leak in GatedDeltaNet cache (#1077) Adam Durham 2026-03-31 23:27:11 -05:00
  • bdeac59767 Inserting logits processors into BatchGenerator in batch_generate (#1008) Arthur Hjorth 2026-04-01 01:48:13 +02:00
  • 6ddfdda1ac Fix SSM dt clamp default for Nemotron-H (#1026) Tarjei Mandt 2026-03-30 22:19:27 +11:00
  • 4d3af3cebc Refactor LRUPromptCache (#1019) Angelos Katharopoulos 2026-03-26 10:04:22 -07:00
  • ed7884cb80 Fix missing cache advance from qwen 3.5 (#1024) Angelos Katharopoulos 2026-03-19 17:20:38 -07:00
  • a1154ab94a Fix the rope mutation in a more natural way rope-mutation Angelos Katharopoulos 2026-03-17 00:43:33 -07:00
  • f8019f7769 Fix flaky test (#1020) Angelos Katharopoulos 2026-03-18 12:51:22 -07:00
  • 564281f793 Supporting delay in mlx_lm benchmark (#1010) AndreasPlt 2026-03-17 00:43:37 +00:00
  • 73c8550478 Nemotron super support (#992) Angelos Katharopoulos 2026-03-16 10:59:14 -07:00
  • ed69f837e6 fall back to ast.literal_eval for malformed JSON in qwen3_coder tool parser (#1004) mm65x 2026-03-16 06:00:51 +00:00
  • cc393b2862 Handle missing content-length header in server (#1001) mm65x 2026-03-16 02:40:29 +00:00
  • 2146e4ed18 avoid mutating input in SuScaledRoPE and YarnRoPE (#1003) mm65x 2026-03-16 01:13:47 +00:00
  • 735a43b275 Delta net precision (#997) Angelos Katharopoulos 2026-03-15 15:39:18 -07:00
  • 332d94ca6f Add allowed-origins to the server (#987) Ryo Ota 2026-03-14 11:22:23 +09:00
  • 480934402d Clear cache trainer memory (#986) n8programs 2026-03-13 21:19:53 -04:00
  • ab157c2d18 Fix test after latest MLX update (#996) Angelos Katharopoulos 2026-03-13 17:07:05 -07:00
  • 5a8ced697e Bump the patch version (#981) Angelos Katharopoulos 2026-03-10 23:27:59 -07:00
  • 760c5abcc8 Fix CompletionsDataset mask_prompt crash (#967) v0.31.1 Eyüp Can Akman 2026-03-11 04:10:06 +03:00
  • b37d7e7f9a Prune models prune-repos Angelos Katharopoulos 2026-02-18 21:46:36 -08:00
  • 43ee5455d3 Move to metal agnostic device_info (#979) Angelos Katharopoulos 2026-03-10 17:41:33 -07:00
  • 23af85703e Late binding caused incorrect cache checkpoint (#976) Angelos Katharopoulos 2026-03-10 13:53:10 -07:00
  • 89c430a9c2 Eval self.left_padding whenever it is updated in BatchRotatingKVCache (#960) rltakashige 2026-03-10 10:14:31 +00:00
  • 4a21ffdf4b Presence and frequency penalties (#971) Angelos Katharopoulos 2026-03-09 22:26:39 -07:00
  • 852119b774 Bump the patch version (#959) Angelos Katharopoulos 2026-03-08 18:04:52 -07:00
  • 044474bc80 Adds tensor parallelism for Qwen 3.5 (#957) v0.31.0 Angelos Katharopoulos 2026-03-06 19:44:53 -08:00
  • 2105aaf9c3 Better caching in the server (#911) Angelos Katharopoulos 2026-03-06 13:42:56 -08:00
  • cff7273a55 Ensure normalization does not promote to fp32 (#951) Angelos Katharopoulos 2026-03-06 13:42:10 -08:00
  • fc7d84448b Bump the version (#954) Angelos Katharopoulos 2026-03-06 13:41:47 -08:00
  • 47be7150a6 fix: convert() uses incorrect defaults for quantization mode (#935) spicyneuron 2026-03-06 09:02:34 +08:00
  • 35fa620279 Add --prefill-step-size as cmd line argument (#943) Yongyue Sun 2026-03-05 09:40:01 +08:00
  • 8162aaad56 step3p5: use rotating cache for sliding attention layers (#949) Noah Lyons 2026-03-04 20:17:29 -05:00
  • 834fac934c fix qwen3.5 sanitize (#928) Awni Hannun 2026-02-24 17:04:43 -08:00
  • 179da774b1 Clear the cache during batch generation (#926) Awni Hannun 2026-02-23 19:50:35 -08:00
  • 720f2369ba add tokens to eval to avoid large graphs when they are not used (#924) Awni Hannun 2026-02-23 14:38:08 -08:00
  • 65725dcec2 Add filter guard to list comprehension (#918) Flynn 2026-02-23 17:22:54 -05:00
  • d4701ba513 clear cache on prompt ingestion in server (#917) n8programs 2026-02-23 15:13:25 -05:00
  • 321e764e0a Make the cache limits more friendly (#910) Angelos Katharopoulos 2026-02-19 13:52:09 -08:00
  • 83ff9c96d5 Improve the cache size limits (#906) Angelos Katharopoulos 2026-02-19 10:13:48 -08:00
  • 9c113f7019 Allow reading LFM2 models nested rope params (#908) Yuri Khrustalev 2026-02-18 19:25:54 -05:00
  • 7d6c5e4af7 Add tie_word_embeddings modulars in mistral and qwen3 moe (#889) Gökdeniz Gülmez 2026-02-18 20:23:40 +01:00
  • ad067ea627 bump for next version (#904) Awni Hannun 2026-02-17 07:39:58 -08:00
  • d7b91e80f0 Fix sharded rms norm in MiniMax M2.5 (#898) Angelos Katharopoulos 2026-02-16 17:20:07 -08:00
  • 1fd521c3c7 fix qwen3.5 casting to fp32 (#902) Awni Hannun 2026-02-16 10:23:31 -08:00
  • 572ada278c server: add usage.prompt_tokens_details.cached_tokens to json response (#849) Ryan Goulden 2026-02-16 08:37:35 -08:00
  • fb47f8fb99 Add the trust remote code option to mlx_lm perplexity (#896) Ivan Fioravanti 2026-02-16 05:43:23 +01:00
  • 7a720882a7 Add JoyAI LLM Flash (#894) Tarjei Mandt 2026-02-16 03:06:44 +11:00
  • 014ebc6a46 Fix mixed quant predicates for MLA models (#892) spicyneuron 2026-02-15 18:44:01 +08:00
  • c6d9d3c9f5 Share model (#871) Angelos Katharopoulos 2026-02-13 15:48:37 -08:00
  • bcf630614f Fix save/load of CacheList (#886) Angelos Katharopoulos 2026-02-12 18:41:48 -08:00
  • 1974376d70 Add GLM5 (#867) v0.30.7 Gökdeniz Gülmez 2026-02-12 19:11:16 +01:00
  • 7e67225e1d Faster DSV32 generation (#885) Tarjei Mandt 2026-02-13 02:35:20 +11:00
  • 0fd3126496 [MODEL] support qwen3.5 series w/o vision (#869) JJJYmmm 2026-02-12 23:23:51 +08:00
  • ca0d1c9630 LongCat MLA (#868) Tarjei Mandt 2026-02-13 01:54:44 +11:00
  • 82edd51a1e Devstral tool parser (#874) Awni Hannun 2026-02-11 09:41:23 -08:00
  • aca4c149a1 Make validation set optional in training process (#857) Gökdeniz Gülmez 2026-02-11 00:24:44 +01:00
  • 8f1c56ec83 Fix DeepSeek V3.2 indexer and weight loading (#866) Tarjei Mandt 2026-02-11 07:15:12 +11:00
  • 84ae19e675 Pythonic tool calling for LFM2 models (#864) viktike 2026-02-10 00:23:07 +01:00
  • 645a326a2e Bump version for next release (#865) Awni Hannun 2026-02-09 14:40:57 -08:00
  • fd6959dca7 Fix Kimi Linear (#853) Tarjei Mandt 2026-02-07 12:31:32 +11:00
  • f18526f8d6 DSV3 MLA (#839) v0.30.6 Awni Hannun 2026-02-04 12:06:42 -08:00
  • 25a4c8369e Fix sliding window mask during generation (#843) Tarjei Mandt 2026-02-05 03:44:01 +11:00
  • e08ec15b72 Fix batch mamba (#842) Awni Hannun 2026-02-03 19:31:42 -08:00
  • b77ec6b951 Fix Step 3.5 Flash model conversion (#840) Tarjei Mandt 2026-02-04 12:03:54 +11:00
  • ab050d1fac Deepseek V3.2 implementation fixes (#838) Sebastian Jug 2026-02-03 10:58:29 -05:00
  • 942b3ed4b6 fix: handle GLM 4.7 tool call fallbacks (#792) Josh Lehman 2026-02-03 06:48:22 -08:00