MOVZXandHinata 5df639bb4e [QSA] Start the clip-counter reader after capture, not on the first forward ...

The FP8 KV clip counter's reader thread used to start on the first eager forward, before CUDA graph capture, so
its device-to-host copies could overlap a capture. _qsa_clip_count now only allocates the per-layer counter; the
reader is started by a hook that guard_telemetry runs after warmup and capture (or at once if capture is already
done). Counting is unchanged. Tests cover the order, idempotence and a failing control that restores the old start.

Co-Authored-By: Hinata <hinata@idihore.id>
Signed-off-by: MOVZX <movzx@yahoo.com>

2026-10-09 20:05:28 +07:00

..

assets

Add pydocstyle to the ruff rules (#52136)

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

benchmarks

[Bugfix][Bench] Clean up synthetic video files and writers (#59529)

2026-10-02 02:51:34 +00:00

compilation

[Feature] Release the CUDA graph pool on sleep (#59160)

2026-10-04 01:13:10 +00:00

config

[PLE] The pull wait/copy op is a graph splitting op

2026-10-09 19:55:22 +07:00

cute_utils

[Bugfix] Support CuTe DSL 4.8.0 block-scale API (#59480)

2026-10-01 02:30:06 +00:00

device_allocator

[Feature] Release the CUDA graph pool on sleep (#59160)

2026-10-04 01:13:10 +00:00

distributed

[Bugfix] Fix FlashInfer all_reduce backend selection (#56891)

2026-10-03 22:18:58 -04:00

engine

[Feature] Release the CUDA graph pool on sleep (#59160)

2026-10-04 01:13:10 +00:00

entrypoints

[Frontend] Name the served models in the model-not-found 404 (#59889)

2026-10-04 20:45:51 +00:00

inputs

[Deprecation] Remove assistant_token_mask support (#57520)

2026-09-19 00:17:51 +08:00

ir

Add pydocstyle to the ruff rules (#52136)

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

kernels

[Refactor] Move auxiliary files out of the repository root (#58572)

2026-09-24 12:38:56 -07:00

logging_utils

[Core][Logging] Add built-in JSON formatter (#58739)

2026-10-01 09:11:46 +01:00

lora

[Kernel][LoRA] Add deterministic split-K=8 LoRA shrink for batch invariance (#59377)

2026-10-04 18:46:16 -04:00

model_executor

ple, tests: debug-gated pull-flag check; diagnostics report that they ran

2026-10-09 20:01:26 +07:00

models

[TopK] In-kernel canonical emission (MERLIN_TOPK_SORTED_EMIT)

2026-10-09 20:04:37 +07:00

multimodal

[Bugfix][Multimodal] Normalize single-channel audio to 1D (#56691)

2026-10-04 06:11:41 +00:00

parser

[Frontend] Constrain non-strict GLM-4.7 tool calls with a shallow structural tag (#56403)

2026-10-02 14:31:53 -04:00

platforms

[Misc][Platform] Check aligned KV block sizes against every attention backend (#58457)

2026-10-04 11:13:20 -04:00

plugins

Add pydocstyle to the ruff rules (#52136)

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

profiler

[Profiler] Unify platform-aware torch profiling (#57460)

2026-09-20 19:17:23 -07:00

ray

Add pydocstyle to the ruff rules (#52136)

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

reasoning

[Agents] Add pre-commit check for agent files and skills (#59657)

2026-10-02 09:38:22 -07:00

renderers

[Frontend] Add output_mode to /inference/v1/generate (RFC #56851 Phase 1) (#58588)

2026-10-03 19:38:25 +08:00

snapshot

[Bugfix] Log CRIU failure details before snapshot cleanup (#59661)

2026-10-03 04:02:27 +00:00

third_party

[GDN] Bound the recurrent state index

2026-10-09 19:43:56 +07:00

tilelang_utils

[Bugfix][ROCm][GLM-5.3-Flash] Apply deferred tilelang.jit already on attribute access (#57192)

2026-09-17 12:02:39 +00:00

tokenizers

[Misc] Remove code paths for Transformers < 5.16.1 (#59762)

2026-10-02 11:17:24 +00:00

tool_parsers

[Frontend] Constrain non-strict GLM-4.7 tool calls with a shallow structural tag (#56403)

2026-10-02 14:31:53 -04:00

tracing

Add pydocstyle to the ruff rules (#52136)

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

transformers_utils

feat: Add video support for the Transformers backend (#57441)

2026-10-05 00:51:15 -07:00

triton_utils

[Feature] Triton kernel dispatcher (#43048)

2026-09-25 04:07:43 +00:00

usage

[Bugfix] Close usage telemetry HTTP sessions (#51219)

2026-08-08 00:29:36 +00:00

utils

[feat] FlashInfer CuteDSL MegaMoE integration (#54049)

2026-10-02 15:43:33 -04:00

v1

[QSA] Start the clip-counter reader after capture, not on the first forward

2026-10-09 20:05:28 +07:00

vllm_flash_attn

[Kernel] Resubmit PR 48666 - Gemma4 FP8 KV FA4 head dim 512 backend selection (#53175)

2026-09-23 20:29:06 -07:00

__init__.py

Add pydocstyle to the ruff rules (#52136)

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

_aiter_ops.py

[ROCm][Perf][GLM-5.3-Flash] Add AITER topk backend for decodes (#58167)

2026-10-02 11:47:41 +01:00

_custom_ops.py

[Attention][MiniMax-M3] NVFP4 KV cache on the MSA sparse attention path (#59300)

2026-10-01 14:39:55 -07:00

_xpu_ops.py

[XPU] Dispatch nn.LayerNorm to fused SYCL kernel via CustomOp (#57172)

2026-09-30 12:19:33 +00:00

collect_env.py

[XPU] fix collecting oneccl version info (#47104)

2026-08-04 11:53:28 +00:00

connections.py

Reject oversized media before fully downloading it (#51896)

2026-08-24 01:15:09 +00:00

env_override.py

[Bugfix] Backport Inductor custom-op pattern matching fix (#58189)

2026-09-22 14:26:49 -07:00

envs.py

[Guards] The bounds-guard telemetry code default is warn

2026-10-09 20:04:39 +07:00

exceptions.py

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

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

forward_context.py

[PCP] Support prefill context parallelism with data parallelism (#57075)

2026-09-23 22:32:36 -04:00

logger.py

[Core][Logging] Add built-in JSON formatter (#58739)

2026-10-01 09:11:46 +01:00

logits_process.py

[Refactor] Remove dead code multiple places (#58002)

2026-09-22 16:36:03 -04:00

logprobs.py

Add pydocstyle to the ruff rules (#52136)

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

model_inspection.py

…

outputs.py

[Feature] Per-row candidate IDs for prefill token scoring (M2 of #56860) (#56984)

2026-10-03 14:17:09 +00:00

pooling_params.py

Add pydocstyle to the ruff rules (#52136)

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

py.typed

…

sampling_params.py

[Feature] Per-row candidate IDs for prefill token scoring (M2 of #56860) (#56984)

2026-10-03 14:17:09 +00:00

scalar_type.py

[Quantization] Enable humming wNaM asymmetric quant (zero_point) with compressed-tensors (#46528)

2026-09-23 10:23:11 -04:00

scripts.py

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

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

sequence.py

…

tasks.py

[Frontend] Move api_server.py out openai folder (#52131)

2026-08-19 09:43:46 +00:00

version.py

[CI] Allowlist-shrink batch 1: wire 17 root-level tests + drop 5 stale watermarking entries into misc.yaml (#58055)

2026-09-27 22:23:46 -07:00