Files

T

MOVZX be95edb7f6 Merge branch 'main' into patches

Signed-off-by: MOVZX <movzx@yahoo.com>

2026-10-16 17:35:42 +07:00

metrics

[Model Runner V2] Support num NaNs in logits (#30187 )

2025-12-09 10:00:49 -08:00

[Perf][Multimodal] Sort multimodal encoder inputs by modality (#60196 )

2026-10-09 10:55:36 +08:00

model_states

Merge branch 'main' into patches

2026-10-15 21:16:39 +07:00

pool

[Bugfix][Pooling] Handle multimodal cache misses without crashing (#58975 )

2026-10-02 10:14:33 -04:00

sample

[watermarking] Add DFlash support to dual-key gumbel-max watermarking (#59554 )

2026-10-15 22:06:06 +00:00

spec_decode

[MRV2][Spec Decode] Quantized draft lm_head for drafters that share the target head (#59973 )

2026-10-15 22:06:40 +00:00

__init__.py

…

async_utils.py

[BugFix][DiffusionGemma][CPU] Copy async output snapshots instead of aliasing reused buffers (#59829 )

2026-10-08 19:16:51 -07:00

attn_utils.py

[XPU] limit FA kernel block size to multipleOf 64 (#57297 )

2026-10-12 10:25:32 +08:00

block_table.py

[KV Cache] GLM-5.3-Flash: use the generic packed KV layout; CircularBufferSpec tail with speculative ring slots (#57169 )

2026-10-09 15:48:37 -04:00

buffer_utils.py

[Bugfix][MRV2] Use int64 index mappings to fix row-offset overflow (#60736 )

2026-10-09 14:38:11 -07:00

cp_utils.py

[MRV2] Miscellaneous code cleanup (#57980 )

2026-09-23 01:51:52 -07:00

cudagraph_utils.py

Merge branch 'main' into patches

2026-10-16 17:35:42 +07:00

dp_utils.py

[Perf] Low-latency CPU all-reduce for DP metadata syncs (#61767 )

2026-10-15 15:29:13 -07:00

ec_connector.py

[MRV2] Miscellaneous code cleanup (#57980 )

2026-09-23 01:51:52 -07:00

eplb_utils.py

[MRV2] Keep warmup out of the EPLB load stats (#56716 )

2026-10-14 00:00:48 +00:00

input_batch.py

[Bugfix][MRV2] Capture DCP FULL cudagraphs with KV context (#61496 )

2026-10-14 19:20:48 +00:00

kv_connector.py

[MRV2] Miscellaneous code structure updates (#60754 )

2026-10-09 17:58:09 -07:00

lora_utils.py

Add pydocstyle to the ruff rules (#52136 )

2026-09-16 16:44:29 +00:00

model_runner.py

[MRV2][Spec Decode] Quantized draft lm_head for drafters that share the target head (#59973 )

2026-10-15 22:06:40 +00:00

pcp_manager.py

[Bugfix][PCP] Align MoE routing padding with gathered tokens (#60946 )

2026-10-14 21:04:00 -04:00

pp_utils.py

[Perf][PP] Skip sampled-token broadcasts whose requests leave the engine (#58542 )

2026-09-30 19:18:55 -07:00

README.md

…

shutdown.py

[MRV2] Add shutdown() method (#41297 )

2026-05-02 21:06:30 -07:00

states.py

[Perf][MRV2] Skip empty draft-token resets on request admission (#61640 )

2026-10-14 12:01:01 -07:00

structured_outputs.py

[Bugfix] Don't let a structured-output request sample from an unmasked row (#54442 )

2026-10-01 02:11:44 +00:00

ubatch_utils.py

[Perf][PP] Skip sampled-token broadcasts whose requests leave the engine (#58542 )

2026-09-30 19:18:55 -07:00

warmup.py

[Spec Decode][MRV2] Warm up top-k/top-p and greedy rejection kernels for speculative decoding (#59260 )

2026-10-14 14:59:21 -04:00

README.md

[Experimental] Model Runner V2

This directory contains the new model runner which is under active development. Ping Woosuk Kwon for any changes.