Julien DenizeandNicolò Lucchesi ce4bdcbda4 [Bugfix] Enable FlashAttention MLA prefill for Mistral Small 4 head dims (#48855) ...

Signed-off-by: juliendenize <julien.denize@mistral.ai>
Co-authored-by: Nicolò Lucchesi <nlucches@redhat.com>

2026-07-17 18:07:00 +08:00

..

attention

[Bugfix] Enable FlashAttention MLA prefill for Mistral Small 4 head dims (#48855)

2026-07-17 18:07:00 +08:00

core

[KV Connector] Fix PD async scheduling race condition for hybrid attn models (#48481)

2026-07-16 11:42:37 +01:00

engine

[rl] Stateful Trainer Send: New Abstractions [1/N] (#48042)

2026-07-17 15:11:06 +08:00

executor

[UX] Log worker exit code when process dies unexpectedly (#38641)

2026-07-07 12:36:29 +00:00

kv_offload

[KV Offload] Add optional tier locality to FS/OBJ KV events (#48281)

2026-07-17 10:31:52 +03:00

metrics

[Refactor] Move iteration logging to the frontend (#46647)

2026-07-15 17:59:05 -07:00

pool

[Core] Move MRV1 late_interaction_runner.py out of MRV2 subtree (#48014)

2026-07-08 18:30:11 +00:00

sample

[Reasoning] Optimize TPOT for thinking budget when used with speculative decoding (#46662)

2026-07-14 18:55:40 +00:00

simple_kv_offload

[Core] Preserve Marconi caching with selective hybrid cache retention (#47782)

2026-07-13 21:24:20 +01:00

spec_decode

[Perf] Optimize clamp to clamp_ (#48143)

2026-07-16 18:41:07 -04:00

structured_output

[Security] Replace diskcache to eliminate pickle deserialization (#44549)

2026-07-14 20:29:24 -07:00

worker

[Bugfix][Multimodal] Fix Qwen3-Omni use_audio_in_video with mixed image/video inputs (#46213)

2026-07-17 08:31:16 +00:00

__init__.py

[V1] AsyncLLM Implementation (#9826)

2024-11-11 23:05:38 +00:00

cudagraph_dispatcher.py

[Model] Add DiffusionGemma Support (#45163)

2026-06-11 22:17:35 -07:00

kv_cache_interface.py

[Bugfix] Include inline per-token-head scales in offloaded page transfer width (#48411)

2026-07-14 16:07:26 +03:00

kv_cache_spec_registry.py

[KVCache] Support Pluggable KVCacheSpec (#37505)

2026-06-03 09:05:16 -07:00

outputs.py

[Core] Cleanup KVConnector handling with PP + fix MRV2 (#43732)

2026-05-28 13:12:03 -07:00

request.py

[Core] Preserve Marconi caching with selective hybrid cache retention (#47782)

2026-07-13 21:24:20 +01:00

serial_utils.py

[Core] Ensure memory is pinned prior to async h2d copy (#45424)

2026-06-20 20:02:24 -07:00

utils.py

[Refactor] Move iteration logging to the frontend (#46647)

2026-07-15 17:59:05 -07:00