3 people 2cf82bcdd1 [Bugfix][DCP] Fix NVIDIA DeepSeek-V3.2 / GLM-5.2 fused attention (#50005) ...

Signed-off-by: 云挚 <ningyunxiao.nyx@antgroup.com>
Signed-off-by: Mikhail Kostryukov <mike@triptrack.net>
Co-authored-by: Mikhail Kostryukov <mike@triptrack.net>
Co-authored-by: Lucas Wilkinson <LucasWilkinson@users.noreply.github.com>

2026-08-31 00:30:45 -07:00

..

assets

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

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

benchmarks

[Bugfix] Restore multimodal support on the plain "vllm" throughput backend (#52168)

2026-08-28 10:28:25 +00:00

compilation

[Model Runner V2] Use MRV2 for all models by default (#53183)

2026-08-27 11:59:34 -07:00

config

[Performance] Optimize Dots3 NOTE runtime (#53517)

2026-08-31 00:05:18 -07:00

cute_utils

[Kimi-K3] Add GEMM-RS for sequence parallelism (#52079)

2026-08-13 09:02:28 -07:00

device_allocator

[Core] Add CuMemAllocator.discard() for tag-selective GPU memory release (#52514)

2026-08-16 23:15:14 +08:00

distributed

[KV Offload] Forward ownership in KV cache events (#52067)

2026-08-31 10:28:41 +03:00

engine

[Model Runner V2] batch-sharded sample (#50465)

2026-08-24 15:49:57 -07:00

entrypoints

[Frontend] Forward cache salt for content parts (#54315)

2026-08-30 23:29:53 -07:00

inputs

[Refactor] Remove InputPreprocessor (#53064)

2026-08-20 08:39:48 +00:00

ir

[CI] Use explicit devices in IR tests (#49513)

2026-07-23 20:08:14 -07:00

kernels

[Bugfix][Triton] Make fp8_min/fp8_max constexpr in _quantize_pad_fp8_kernel (#51872)

2026-08-12 03:09:54 +00:00

logging_utils

…

lora

[LoRA] Cleanup VocabParallelEmbedding (#53843)

2026-08-26 16:19:35 +08:00

model_executor

[Kernel] Retire the DSv3 router GEMM CUDA kernel (#54040)

2026-08-31 07:15:31 +00:00

models

[Bugfix][DCP] Fix NVIDIA DeepSeek-V3.2 / GLM-5.2 fused attention (#50005)

2026-08-31 00:30:45 -07:00

multimodal

[Bugfix][Multimodal] Honor modality-scoped mm_processor_kwargs (#53808)

2026-08-31 06:37:12 +00:00

parser

[Hy4] support Hy4-preview model (#54160)

2026-08-29 12:07:28 +08:00

platforms

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

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

plugins

[MISC] Cleanup deprecated parameters (#53559)

2026-08-24 10:19:04 -07:00

profiler

[Profiler] Fix start_profile permanently no-op after max_iterations auto-stop (#51839)

2026-08-25 08:13:29 -04:00

ray

Propagate Flash Attention cache configuration to Ray workers (#49177)

2026-07-21 07:47:53 +00:00

reasoning

[Hy4] support Hy4-preview model (#54160)

2026-08-29 12:07:28 +08:00

renderers

[Bugfix][Frontend] Truncate prompt_token_offsets with the prompt (#54407)

2026-08-31 06:13:37 +00:00

third_party

[Bugfix][Kernel] Keep packed GDN decode beta in FP32 (#53877)

2026-08-30 22:15:36 +08:00

tilelang_utils

[ROCm] Defer tilelang import through its import from vllm.tilelang_utils import tilelang and relaxed has_tilelang (#51159)

2026-08-13 13:12:40 -05:00

tokenizers

[Bugfix][Tokenizer] Replace bare asserts in the DeepSeek V4 encoder (#53747)

2026-08-25 16:30:23 +00:00

tool_parsers

[Doc] Fix griffe warnings in HYV4 tool parser (#54412)

2026-08-30 13:21:22 +00:00

tracing

…

transformers_utils

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

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

triton_utils

Exclude the cpu backend from vLLM's active-Triton-driver count (#53530) (#53530)

2026-08-25 04:13:13 +00:00

usage

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

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

utils

[Bugfix] Fix ncclCommQueryProperties heap overflow with NCCL >= 2.31 (#53008)

2026-08-29 00:25:00 +00:00

v1

[KV Offload] Forward ownership in KV cache events (#52067)

2026-08-31 10:28:41 +03:00

vllm_flash_attn

Revert "[Kernel] Gemma-4 FA4 FP8 Kernel" (#52987)

2026-08-19 10:48:46 -07:00

__init__.py

…

_aiter_ops.py

[ROCm] remove VLLM_ROCM_USE_AITER_FP4_ASM_GEMM environment variable; make w4a4 use the preshuffle triton+asm by default (#53141)

2026-08-28 18:59:53 +00:00

_custom_ops.py

[Kernel] Retire the DSv3 router GEMM CUDA kernel (#54040)

2026-08-31 07:15:31 +00:00

_xpu_ops.py

[Bugfix][XPU] Run GDN attention as eager break under breakable cudagraph (#51928)

2026-08-12 16:23:39 +08:00

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

…

envs.py

[Hy4] support Hy4-preview model (#54160)

2026-08-29 12:07:28 +08:00

exceptions.py

[Frontend][Core] Standardize request error handling with VLLMError hierarchy (#49665)

2026-07-29 04:20:40 +00:00

forward_context.py

[Feature] Support sequence parallel without the need for DP, 1.9%~5.0% E2E Throughput Improvement (#47070)

2026-07-05 05:41:30 -07:00

logger.py

…

logits_process.py

…

logprobs.py

[Bugfix] Handle empty FlatLogprobs slices and delta output (#53704)

2026-08-26 21:00:54 +08:00

model_inspection.py

…

outputs.py

[Frontend][Core][Spec Decode] Per-request acceptance stats in OpenAI API responses (#48915)

2026-08-20 15:42:49 -07:00

pooling_params.py

[MISC] Cleanup deprecated parameters (#53559)

2026-08-24 10:19:04 -07:00

py.typed

…

sampling_params.py

[Bugfix][Frontend] Validate stop_token_ids against vocab size (#54196)

2026-08-29 16:25:35 +00:00

scalar_type.py

Humming support for 2/3/5/6/7-bit pack-quantized weight-only inference (#46389)

2026-06-24 13:53:54 -04:00

scripts.py

…

sequence.py

…

tasks.py

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

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

version.py

…