Lehua Ding e184c9c510 [perf] Use CPU tensor to reduce GPU->CPU sync (#25884) ...

Signed-off-by: Lehua Ding <lehuading@tencent.com>

2025-09-30 19:51:16 +08:00

..

attention

[New Model] DeepSeek-V3.2 (Rebased to Main) (#25896)

2025-09-30 17:14:41 +08:00

core

[New Model] DeepSeek-V3.2 (Rebased to Main) (#25896)

2025-09-30 17:14:41 +08:00

engine

[Core] GC Debug callback (#24829)

2025-09-27 17:53:31 +00:00

executor

[V0 Deprecation] Remove vllm.worker and update according imports (#25901)

2025-09-29 23:26:11 +00:00

kv_offload

[Docs] Enable fail_on_warning for the docs build in CI (#25580)

2025-09-24 19:30:33 +00:00

metrics

[Core] Don't count preempted tokens in prefix cache hit rate (#25787)

2025-09-27 00:16:40 +00:00

pool

[Performance] V1 Pooling Models E2E Performance Optimization (#23162)

2025-08-21 13:26:09 +00:00

sample

[Spec Decode] Add Batch Parallel Ngram. Upto 8x lower overhead. (#24986)

2025-09-25 15:22:03 -07:00

spec_decode

[New Model] DeepSeek-V3.2 (Rebased to Main) (#25896)

2025-09-30 17:14:41 +08:00

structured_output

[Build] Update Xgrammar to 0.1.25 (#25467)

2025-09-23 21:25:46 +00:00

worker

[perf] Use CPU tensor to reduce GPU->CPU sync (#25884)

2025-09-30 19:51:16 +08:00

__init__.py

[V1] AsyncLLM Implementation (#9826)

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

cudagraph_dispatcher.py

Remove redundant cudagraph dispatcher warning (#25841)

2025-09-28 17:12:42 -04:00

kv_cache_interface.py

[New Model] DeepSeek-V3.2 (Rebased to Main) (#25896)

2025-09-30 17:14:41 +08:00

outputs.py

[Misc] Simplify PoolerOutput and move to v1/outputs (#25629)

2025-09-25 06:47:03 +00:00

request.py

[Core] Don't count preempted tokens in prefix cache hit rate (#25787)

2025-09-27 00:16:40 +00:00

serial_utils.py

fix: revert cast to cpu in MsgpackEncoder._encode_tensor to avoid hidden performance regressions (#25738)

2025-09-26 01:18:05 -07:00

utils.py

[Bugfix] Optimize CpuGpuBuffer initialization (#25447)

2025-09-27 08:07:36 +08:00