EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

[Misc][V0 Deprecation] Remove multi-step worker (#1809)

Browse Source 
### What this PR does / why we need it?
Remove multi-step worker

This PR is a part of
https://github.com/vllm-project/vllm-ascend/issues/1620.

- vLLM version: v0.9.2
- vLLM main:
235bfd5dfe

---------

Signed-off-by: shen-shanshan <467638484@qq.com>
This commit is contained in:
Shanshan Shen
2025-07-15 19:48:47 +08:00
committed by GitHub
parent bf2549856f
commit a929699e98
4 changed files with 0 additions and 303 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
vllm_ascend/patch/__init__.py
View File

@@ -73,23 +73,6 @@
# Future Plan:
# Keep this patch in vllm-ascend.
#
# ** File: worker/patch_common/patch_multi_step_worker.py **
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# 1. `vllm.spec_decode.multi_step_worker.MultiStepWorker.sampler_output`
# Why:
# There are cuda hard code (current_platform.is_cuda_alike()) in
# `MultiStepWorker.sampler_output`, and we need to use the patched `TP1DraftModelRunner` in it.
# How：
# Make speculative decoding extensible to different backends.
# - support attention metadata register to the set supported spec decode
# - offer a api in platform to determine whether spec decode is supported,
# and deprecate is_cuda_alike in it.
# Related PR (if no, explain why):
# - https://github.com/vllm-project/vllm/pull/15195
# - https://github.com/vllm-project/vllm-ascend/pull/395
# Future Plan:
# Revert it when the related pr is merged in vllm and vllm-ascend.
#
# ** File: worker/patch_common/patch_spec_decode_worker.py **
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# 1. `vllm.spec_decode.spec_decode_worker.SpecDecodeWorker.create_worker`