Woosuk KwonandClaude 50cfbbca7a [MRV2] Draw 64-bit uniforms for fp32 Gumbel sampling ...

tl.rand only resolves the uniform down to 2**-31, which hard-caps the
fp32 Gumbel noise at ~21.5 and coarsely buckets the argmax-deciding
tail. Philox already generates 128 bits per call (tl.rand discards 96),
so draw 64 bits instead and convert uint64 -> fp32 directly: the tail
now resolves down to 2**-64 (noise up to ~44.4) with no fp64 arithmetic
and no measurable cost. The fp64 path is unchanged.

Co-authored-by: Claude
Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>

2026-07-03 06:22:59 +00:00

..

basic_correctness

[Platform] Replace torch.cuda.mem_get_info with torch.accelerator.get_memory_info (#44825)

2026-06-30 14:39:52 +08:00

benchmarks

[Bench][BugFix] Fix empty decoder prompt for Cohere ASR in throughput benchmark (#47135)

2026-07-01 15:42:27 +00:00

compile

[ROCm][MLA] Fuse MLA q/kv RMSNorm + FP8 per-token quant in the FP8 attention path (#44977)

2026-07-02 13:00:41 +08:00

config

[ROCm][CI] Fix nixl tests (#45219)

2026-06-23 13:11:40 -05:00

cuda

[Hardware] Replace torch.cuda.device_count/current_device/set_device API (#36145)

2026-03-12 07:57:47 -07:00

detokenizer

[CI/Build] Remove use of skip_v1 (#34699)

2026-02-18 12:19:11 +08:00

distributed

[ModelRunner V2] Enable by default for all dense models (#44443)

2026-07-02 18:48:57 +08:00

engine

[Feat] Add runtime monitor for post-warmup CuTeDSL compilation (#46167)

2026-06-23 23:33:17 +00:00

entrypoints

[Test] Run SageMaker handler-override tests in-process via TestClient (#47250)

2026-07-01 09:14:00 +00:00

evals

[CI] Move distributed small LM eval to B200 (#47048)

2026-06-30 13:34:25 -04:00

fusion

Add the QuantizedActivation linear-kernel contract (#44260)

2026-06-12 13:48:15 -07:00

ir

[Perf] Set IR Op Priority Once at Worker Init (#42631)

2026-05-15 15:56:13 +00:00

kernels

Xqa decode kernels (#43232)

2026-07-02 12:32:05 -07:00

lora

[CI/Build] Fix LoRA testing (#47242)

2026-07-01 15:36:13 +08:00

model_executor

[XPU][CI] Enable shared loader test (#45977)

2026-06-30 11:20:33 +00:00

models

Add Laguna XS.2.1 DFlash drafter support (#46853)

2026-07-02 18:09:27 -07:00

multimodal

[Bugfix] Align OpenCV video metadata timeline (#47099)

2026-06-30 20:43:42 +00:00

parser

[Bugfix][Frontend][gpt-oss] Return raw output when Harmony parser ends non-terminal (#47062)

2026-07-01 01:46:01 +00:00

plugins

[Misc][Model] add io processor for query/document embeddings from ColBERT (jinaai/jina-colbert-v2) (#45210)

2026-06-16 01:32:32 +00:00

plugins_tests

[BugFix][CI] Fix scheduler plugin test (#45897)

2026-06-17 06:30:49 +00:00

prompts

[BugFix] Fix input positions for long context with sliding window (#2088)

2023-12-13 12:28:13 -08:00

quantization

Add MiniMax-M3 modelopt nvfp4 support (#46756)

2026-06-29 09:29:39 -07:00

reasoning

[Frontend] Add Streaming Parser Engine and new Kimi k2.5/k2.6/k2.7 Parser (#46610)

2026-06-30 07:53:17 +00:00

renderers

fix(security): bound tokenizer work when explicit truncation_side is set (#47007)

2026-06-30 23:08:51 +08:00

rocm/aiter

[ROCm] Fix AITER FP8 quantization schema tests (#46414)

2026-06-24 22:29:19 +08:00

samplers

[Bugfix] Fix beam search candidate indexing when logprobs count varies (#47126)

2026-07-01 07:07:06 +00:00

spec_decode

[Feature] Support custom callable proposer backend for speculative decoding (#39487)

2026-05-13 16:53:01 +00:00

standalone_tests

[ROCm][CI] pass merge-base to container for python-only wheel metadata (#45869)

2026-06-23 15:44:43 +00:00

system_messages

[V1] Implement Cascade Attention (#11635)

2025-01-01 21:56:46 +09:00

tokenizers_

Remove grok model arch from vllm (#46706)

2026-06-25 23:02:10 -07:00

tool_parsers

[Bugfix][Tool Parser] poolside_v1: accept tool calls without newline after function name (#47311)

2026-07-02 12:08:37 -07:00

tool_use

[Parser][Bugfix] Ensure tool call or other special tokens don't leak in non-streaming tool parsing (#46875)

2026-06-30 13:46:53 -04:00

tools

[ROCm][CI] Adding metadata (#47477)

2026-07-02 17:45:23 -07:00

tracing

[CI][Tracing] Fix race condition by adding server readiness check (#34364)

2026-02-16 12:57:39 +00:00

transformers_utils

[CI] Don't try and download files that we already know don't exist (#46854)

2026-06-26 23:56:39 +00:00

utils_

[Platform] Replace torch.cuda.mem_get_info with torch.accelerator.get_memory_info (#44825)

2026-06-30 14:39:52 +08:00

v1

[MRV2] Draw 64-bit uniforms for fp32 Gumbel sampling

2026-07-03 06:22:59 +00:00

vllm_test_utils

Update Optional[x] -> x | None and Union[x, y] to x | y (#26633)

2025-10-12 09:51:31 -07:00

weight_loading

[Quant] Consolidate GPTQ: rename gptq_marlin.py to auto_gptq.py (#38288)

2026-05-15 08:25:52 +08:00

__init__.py

[Small] Formatter only checks lints in changed files (#1528)

2023-10-31 15:39:38 -07:00

ci_envs.py

[Model][0/N] Improve all pooling task | clean up (#25817)

2025-10-13 16:44:50 +08:00

conftest.py

[ROCm][CI] Explicitly tear down multimodal offline LLMs (#46999)

2026-06-29 07:59:24 +00:00

test_access_log_filter.py

[Logging] add --disable-access-log-for-endpoints CLI option (#30011)

2026-01-26 21:49:03 +00:00

test_attention_backend_registry.py

[1/N][Attention] Restructure attention: move files (#31916)

2026-01-09 13:10:24 -08:00

test_config.py

[ModelRunner V2] Enable by default for all dense models (#44443)

2026-07-02 18:48:57 +08:00

test_embedded_commit.py

Convert formatting to use ruff instead of yapf + isort (#26247)

2025-10-05 07:06:22 -07:00

test_envs.py

[Build] fix self-contradictory precompiled-flag orthogonality test (#44942)

2026-06-11 12:49:08 +08:00

test_force_first_config.py

[Perf] Add VLLM_TRITON_FORCE_FIRST_CONFIG to skip Triton autotuning (#42425)

2026-06-16 15:16:45 +02:00

test_fxgraphcache_pickle_patch.py

[ZenCPU] Make PT Backport Patch Accessible to vLLM (#38205)

2026-04-10 08:29:35 +00:00

test_inputs.py

[Refactor] Pass full VllmConfig to Renderer (#34485)

2026-02-12 22:48:38 -08:00

test_jit_monitor.py

[Feat] Add runtime monitor for post-warmup CuTeDSL compilation (#46167)

2026-06-23 23:33:17 +00:00

test_logger.py

[Frontend] Consolidate online serving utils. (#44479)

2026-06-04 06:49:31 +00:00

test_logprobs.py

[Core] Switch Flat logprob control from environment variable to SamplingParams (#28914)

2025-11-19 02:10:02 +00:00

test_outputs.py

Convert formatting to use ruff instead of yapf + isort (#26247)

2025-10-05 07:06:22 -07:00

test_pooling_params.py

[Bugfix] Reject matryoshka embedding dimensions above hidden size (#46313)

2026-06-22 10:16:35 +00:00

test_ray_env_utils.py

[Feat][Executor] Introduce RayExecutorV2 (#36836)

2026-04-01 14:34:29 -07:00

test_ray_env.py

fix nixl side-channel host selection (#41806)

2026-05-11 07:40:37 +00:00

test_regression.py

[CI/Build] Use modelscope's international site for regression test (#42176)

2026-05-09 19:47:09 -07:00

test_scalartype.py

Convert formatting to use ruff instead of yapf + isort (#26247)

2025-10-05 07:06:22 -07:00

test_sequence.py

Convert formatting to use ruff instead of yapf + isort (#26247)

2025-10-05 07:06:22 -07:00

test_triton_utils.py

Convert formatting to use ruff instead of yapf + isort (#26247)

2025-10-05 07:06:22 -07:00

test_version.py

Convert formatting to use ruff instead of yapf + isort (#26247)

2025-10-05 07:06:22 -07:00

test_vllm_port.py

Convert formatting to use ruff instead of yapf + isort (#26247)

2025-10-05 07:06:22 -07:00

test_zen_cpu_platform_detection.py

In-Tree AMD Zen CPU Backend via zentorch [1/N] (#35970)

2026-03-15 23:35:35 +00:00

utils.py

[XPU] Optimize XPU worker shutdown logic to prevent resource leak (#46433)

2026-06-30 15:27:21 +08:00