828 Commits

Author SHA1 Message Date
Arun Raj df1d3f3c9a Fix Gemma 4 sanitize() not stripping KV projections for shared layers (#1240) 2026-05-04 15:26:18 -07:00
Angelos Katharopoulos ed1fca4cef Thread local generation stream (#1090) v0.31.3 2026-04-22 00:34:09 -07:00
glyphVault 4f5cbd2a4f Fix Gemma 4 KV-shared layers creating unused projections (#1158) 2026-04-21 16:44:13 -07:00
Angelos Katharopoulos 3cd9a52df2 Fix ArraysCache extend (#1177) 2026-04-21 16:41:49 -07:00
Eyüp Can Akman 2f1ab85aec Fix Mistral empty tool_call_end flipping state machine to normal (#1151) 2026-04-21 01:36:56 -07:00
AkashKhamkar f3bb10c141 Fix Gemma4 tool parser: support hyphenated names and braces in strings (#1150) 2026-04-21 01:15:58 -07:00
Sherry Lo e1c24b3237 fix: handle NoneType check for think tokens in TokenizerWrapper (#1167) 2026-04-21 01:13:15 -07:00
Luis Molina f39cb8e934 Fix dwq: check for actual safetensors in target_dir (#1173) 2026-04-21 00:41:53 -07:00
TechToboggan a9856b485d Fix batch dimension mismatch in ArraysCache extend() (#1169)
Co-authored-by: Tristan Stahnke <tristan@melchior.lan>
Co-authored-by: Angelos Katharopoulos <a_katharopoulos@apple.com>
2026-04-20 23:42:03 -07:00
Andrei Panferov e92138cb01 Apertus tie_word_embeddings fix (#1143) 2026-04-20 23:47:40 +02:00
Siiea-ai a401730941 Fix missing tree_reduce import in models/cache.py (#1165) 2026-04-20 11:31:58 -07:00
Tarjei Mandt 6d114686e5 Fix MiniMax M2 parallel tool calling (#1171) 2026-04-20 10:57:20 -07:00
Tarjei Mandt aa4f880fb3 Fix parallel tool call handling in server (#1170) 2026-04-19 23:58:25 -07:00
razorback16 62f38aeb51 Fix batch dimension mismatch in BatchKVCache and BatchRotatingKVCache extend() (#1141) 2026-04-14 17:21:31 -07:00
Angelos Katharopoulos d9c63fff67 Bump the patch version (#1124) 2026-04-08 02:04:27 -07:00
Neil Mehta dcbf6e33d1 Align batch logits processor token contract (#1115) v0.31.2 2026-04-06 18:07:35 -07:00
Angelos Katharopoulos f26fddfd3b Gemma4 final fixes and multi-token think/tool start/end (#1114) 2026-04-06 17:40:48 -07:00
Tarjei Mandt f56d99712c Fix output corruption in speculative decoding (#1109) 2026-04-06 16:00:38 -07:00
spicyneuron c65c27b450 Fix Gemma 4 quantized per-layer projection loading (#1112) 2026-04-06 13:26:07 -07:00
Nic Davidson 3257c3df17 Add Gemma 4 tool call parser (#1105) 2026-04-04 17:21:45 -07:00
Angelos Katharopoulos d4eb136d44 Bring back max-kv-size to the batch generator (#1106) 2026-04-04 17:12:43 -07:00
Prince Canuma 4469ad4647 Add gemma 4 (#1093)
Co-authored-by: N8 <n8@n8programs.com>
Co-authored-by: Angelos Katharopoulos <a_katharopoulos@apple.com>
2026-04-04 04:47:03 -07:00
Matteo Celani f79dba7832 perf: use max instead of argsort in apply_min_p sampling (#1083) 2026-04-01 16:26:59 -07:00
Angelos Katharopoulos 3f9d179fd1 Batch generation refactoring and various fixes (#1072) 2026-04-01 15:07:50 -07:00
Lik Xun Yuan (Lx) 9dc023beed Fix PromptTrie.pop_prefixes() off-by-one when pruning immediate prefixes (#1078)
Signed-off-by: Yuan Lik Xun <lxyuan0420@gmail.com>
Co-authored-by: Angelos Katharopoulos <a_katharopoulos@apple.com>
2026-04-01 11:51:19 -07:00
Adam Durham 9dcefa5272 fix: break shared-buffer memory leak in GatedDeltaNet cache (#1077)
Co-authored-by: Angelos Katharopoulos <a_katharopoulos@apple.com>
2026-03-31 21:27:11 -07:00
Arthur Hjorth bdeac59767 Inserting logits processors into BatchGenerator in batch_generate (#1008)
Co-authored-by: Angelos Katharopoulos <a_katharopoulos@apple.com>
2026-03-31 16:48:13 -07:00
Tarjei Mandt 6ddfdda1ac Fix SSM dt clamp default for Nemotron-H (#1026) 2026-03-30 04:19:27 -07:00
Angelos Katharopoulos 4d3af3cebc Refactor LRUPromptCache (#1019) 2026-03-26 10:04:22 -07:00
Angelos Katharopoulos ed7884cb80 Fix missing cache advance from qwen 3.5 (#1024) 2026-03-19 17:20:38 -07:00
Angelos Katharopoulos f8019f7769 Fix flaky test (#1020) 2026-03-18 12:51:22 -07:00
AndreasPlt 564281f793 Supporting delay in mlx_lm benchmark (#1010) 2026-03-16 17:43:37 -07:00
Angelos Katharopoulos 73c8550478 Nemotron super support (#992) 2026-03-16 10:59:14 -07:00
mm65x ed69f837e6 fall back to ast.literal_eval for malformed JSON in qwen3_coder tool parser (#1004) 2026-03-15 23:00:51 -07:00
mm65x cc393b2862 Handle missing content-length header in server (#1001) 2026-03-15 19:40:29 -07:00
mm65x 2146e4ed18 avoid mutating input in SuScaledRoPE and YarnRoPE (#1003) 2026-03-15 18:13:47 -07:00
Angelos Katharopoulos 735a43b275 Delta net precision (#997) 2026-03-15 15:39:18 -07:00
Ryo Ota 332d94ca6f Add allowed-origins to the server (#987) 2026-03-13 19:22:23 -07:00
n8programs 480934402d Clear cache trainer memory (#986)
Co-authored-by: N8 <n8@n8programs.com>
Co-authored-by: Angelos Katharopoulos <a_katharopoulos@apple.com>
2026-03-13 18:19:53 -07:00
Angelos Katharopoulos ab157c2d18 Fix test after latest MLX update (#996) 2026-03-13 17:07:05 -07:00
Angelos Katharopoulos 5a8ced697e Bump the patch version (#981) 2026-03-10 23:27:59 -07:00
Eyüp Can Akman 760c5abcc8 Fix CompletionsDataset mask_prompt crash (#967) v0.31.1 2026-03-10 18:10:06 -07:00
Angelos Katharopoulos 43ee5455d3 Move to metal agnostic device_info (#979) 2026-03-10 17:41:33 -07:00
Angelos Katharopoulos 23af85703e Late binding caused incorrect cache checkpoint (#976) 2026-03-10 13:53:10 -07:00
rltakashige 89c430a9c2 Eval self.left_padding whenever it is updated in BatchRotatingKVCache (#960)
Co-authored-by: Angelos Katharopoulos <a_katharopoulos@apple.com>
2026-03-10 03:14:31 -07:00
Angelos Katharopoulos 4a21ffdf4b Presence and frequency penalties (#971) 2026-03-09 22:26:39 -07:00
Angelos Katharopoulos 852119b774 Bump the patch version (#959) 2026-03-08 18:04:52 -07:00
Angelos Katharopoulos 044474bc80 Adds tensor parallelism for Qwen 3.5 (#957) v0.31.0 2026-03-06 19:44:53 -08:00
Angelos Katharopoulos 2105aaf9c3 Better caching in the server (#911) 2026-03-06 13:42:56 -08:00
Angelos Katharopoulos cff7273a55 Ensure normalization does not promote to fp32 (#951) 2026-03-06 13:42:10 -08:00