Files

T

560ef78bfe [Perf][Model Runner V2] Compact sampling masks on GPU instead of unpacking the full-vocab bitmask on CPU (#54901 )

Signed-off-by: aoshen02 <aoshen@inferact.ai>
Signed-off-by: vx120 <893600387@qq.com>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: vx120 <893600387@qq.com>

2026-09-03 19:03:36 -07:00

metrics

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

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

[Perf] Avoid h2d copies from non-pinned CPU tensors (#54299 )

2026-08-29 12:06:30 +08:00

model_states

[Model] add GLM-5.3-Flash support (#53906 )

2026-09-03 09:40:35 -07:00

pool

[Pooling] Improve BGE-M3 sync pooling throughput by up to 3.13% (#53464 )

2026-08-24 08:33:32 -04:00

sample

[Perf][Model Runner V2] Compact sampling masks on GPU instead of unpacking the full-vocab bitmask on CPU (#54901 )

2026-09-03 19:03:36 -07:00

spec_decode

[Model Runner V2][Spec Decode] Skip DP sync for all speculator uniform decodes (#54856 )

2026-09-03 10:32:33 +00:00

__init__.py

GPU Model Runner V2 (#25266 )

2025-11-21 08:20:55 -08:00

async_utils.py

[Perf][Model Runner V2] Compact sampling masks on GPU instead of unpacking the full-vocab bitmask on CPU (#54901 )

2026-09-03 19:03:36 -07:00

attn_utils.py

[Mamba] Add FlashInfer ReplaySSM backend (#52506 )

2026-09-01 10:26:43 -04:00

block_table.py

[Model] Support Qwen3.8-Flash-Next (#53896 )

2026-08-31 05:57:56 +00:00

buffer_utils.py

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

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

cp_utils.py

[Spec decode] Support Kimi-K3 DCP with DSpark (#52188 )

2026-08-17 13:08:45 -07:00

cudagraph_utils.py

[Bugfix] Raise for unavailable piecewise CUDA graphs (#54782 )

2026-09-02 08:52:24 +00:00

dp_utils.py

[Model Runner V2][Spec Decode] Skip DP sync for all speculator uniform decodes (#54856 )

2026-09-03 10:32:33 +00:00

ec_connector.py

[EC Connector] Added Build Connector Worker Meta for EC Connector (#49585 )

2026-08-15 15:40:38 -07:00

eplb_utils.py

[Elastic EP] Reduce eager-mode reconfiguration downtime (#51885 )

2026-08-19 02:13:06 +00:00

input_batch.py

[Bugfix][PP] Never drop a decoding request from the sampled-token broadcast (#54436 )

2026-08-31 23:49:54 +00:00

kv_connector.py

[Core][KV Connector] Start async KV loads after the forward launch when no sync loads are scheduled (#53333 )

2026-08-28 21:58:00 +00:00

lora_utils.py

[BugFix][MRV2] Don't create dummy requests longer than max_model_len (#49751 )

2026-07-27 02:04:20 +00:00

model_runner.py

[Model] add GLM-5.3-Flash support (#53906 )

2026-09-03 09:40:35 -07:00

pcp_manager.py

[Bugfix][PP] Never drop a decoding request from the sampled-token broadcast (#54436 )

2026-08-31 23:49:54 +00:00

pp_utils.py

[Bugfix][PP] Never drop a decoding request from the sampled-token broadcast (#54436 )

2026-08-31 23:49:54 +00:00

README.md

GPU Model Runner V2 (#25266 )

2025-11-21 08:20:55 -08:00

shutdown.py

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

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

states.py

[Model Runner V2][Spec Decode] Add multi-layer MTP speculator (#48892 )

2026-07-30 15:54:15 -07:00

structured_outputs.py

[Bugfix][Spec Decode][Structured Output] DSpark: fix the grammar bitmask mapping when the draft budget is zero (#52436 )

2026-08-16 11:09:23 +00:00

warmup.py

[Model] Support Qwen3.8-Flash-Next (#53896 )

2026-08-31 05:57:56 +00:00

README.md

[Experimental] Model Runner V2

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