Wang Kunpeng 13c4a9c78b [bugfix]Fix accuracy issue in PCP/DCP with speculative decoding (#6491) ...

### What this PR does / why we need it?

This PR fixes an accuracy issue that occurs when using Prefill/Decode
Context Parallelism (PCP/DCP) in conjunction with speculative decoding
(MTP). The issue is caused by an irregular attention mask shape when
both features are enabled.

The fix involves flattening the `block_table` for speculative decoding
requests under PCP/DCP to ensure a regular attention mask. This PR also
introduces a `use_cp` property for cleaner code and updates dummy runs
to handle this scenario correctly.

### Does this PR introduce _any_ user-facing change?

No. This is a bug fix that improves accuracy and should not have
user-facing API changes.

### How was this patch tested?

- vLLM version: v0.15.0
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.15.0

---------

Signed-off-by: Wang Kunpeng <1289706727@qq.com>

2026-02-05 10:06:14 +08:00

..

_310p

[Refactor] MLP weight prefetch to consistency with MoE Model's prefetching in terms of code and usage (#6442)

2026-02-04 09:08:18 +08:00

_cann_ops_custom

[Kernel] add custom op GmmSwigluQuantWeightNzTensorList (#3804)

2025-11-28 18:06:39 +08:00

attention

[Bugfix]Fix the compatibility issue of may_reinitialize_input_batch (#6290)

2026-02-02 19:16:26 +08:00

compilation

[Fusion] Add rmsnorm dynamic quant fusion pass (#6274)

2026-02-04 15:53:53 +08:00

core

[0.14.1][bugfix][sched] fix incompatibility of RecomputeScheduler with vllm v0.14.1 (#6286)

2026-01-28 20:16:58 +08:00

device

[Lint]Style: Convert vllm-ascend/ to ruff format(Batch #2) (#5977)

2026-01-19 08:59:46 +08:00

device_allocator

[Lint]Style: Convert vllm-ascend/ to ruff format(Batch #2) (#5977)

2026-01-19 08:59:46 +08:00

distributed

[Bugfix]Fix of Pooling Code and Update of Pooling Usage Guide (#6126)

2026-02-04 16:35:41 +08:00

eplb

[EPLB][Bugfix] EPLB support fp/bf16 (#5531)

2026-01-26 14:28:16 +08:00

kv_offload

[Lint]Style: Convert vllm-ascend/ to ruff format(Batch #5) (#5996)

2026-01-24 22:45:38 +08:00

lora

[Lint]Style: Convert vllm-ascend/ to ruff format(Batch #5) (#5996)

2026-01-24 22:45:38 +08:00

model_loader

[Lint]Style: Convert vllm-ascend/ to ruff format(Batch #6) (#6001)

2026-01-24 22:08:33 +08:00

ops

perf: adaptive block size selection in linear_persistent kernel (#6537)

2026-02-04 21:36:26 +08:00

patch

[Bugfix][CI]Add qwen3Next MTP+Full Decode (#6047)

2026-02-03 14:26:21 +08:00

quantization

[Quant] GLM4.7-Flash Support W8A8 (#6492)

2026-02-03 19:49:58 +08:00

sample

[ops] support advanced apply_top_k_top_p without top_k constraint (#6098)

2026-01-26 09:08:42 +08:00

spec_decode

[Refactor][EAGLE] 6/N route mtp to eagle except pcp/dcp+mtp (#6349)

2026-02-02 19:15:31 +08:00

worker

[bugfix]Fix accuracy issue in PCP/DCP with speculative decoding (#6491)

2026-02-05 10:06:14 +08:00

xlite

[CI] optimize lint term (#5986)

2026-01-22 15:46:59 +08:00

__init__.py

[Lint]Style: Convert vllm-ascend/compilation to ruff format (#5912)

2026-01-16 20:57:46 +08:00

ascend_config.py

[Refactor] MLP weight prefetch to consistency with MoE Model's prefetching in terms of code and usage (#6442)

2026-02-04 09:08:18 +08:00

ascend_forward_context.py

[Refactor] MLP weight prefetch to consistency with MoE Model's prefetching in terms of code and usage (#6442)

2026-02-04 09:08:18 +08:00

batch_invariant.py

[Lint]Style: Convert vllm-ascend/ to ruff format(Batch #2) (#5977)

2026-01-19 08:59:46 +08:00

cpu_binding.py

[Lint]Style: Convert vllm-ascend/compilation to ruff format (#5912)

2026-01-16 20:57:46 +08:00

envs.py

[Kernel] Add AscendC fused op transpose_kv_cache_by_block to speed up GQA transfer (#6366)

2026-02-03 14:10:01 +08:00

flash_common3_context.py

[Lint]Style: Convert vllm-ascend/compilation to ruff format (#5912)

2026-01-16 20:57:46 +08:00

meta_registration.py

[Lint]Style: Convert vllm-ascend/compilation to ruff format (#5912)

2026-01-16 20:57:46 +08:00

platform.py

[Feat.]: support 310p w8a8 (#6454)

2026-02-03 14:13:06 +08:00

profiling_config.py

[Core][Misc] Clean up ProfileExecuteDuration (#6461)

2026-02-01 20:06:01 +08:00

utils.py

[Refact.]: Refactor some leftover implementations of 300I DUO in the main branch. (#6425)

2026-02-02 16:12:04 +08:00