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

Clean up v0.9.1 code (#1672)

Browse Source 
vllm has released 0.9.2. This PR drop 0.9.1 support.

- vLLM version: v0.9.1
- vLLM main:
b942c094e3

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-07-09 08:52:24 +08:00
committed by GitHub
parent 0d4bc03946
commit 830332ebfc
23 changed files with 205 additions and 846 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
vllm_ascend/patch/__init__.py
View File

@@ -24,9 +24,9 @@
# each worker's `__init__` function.
#
# Then in each kind of patch, there are three folders:
# - patch_0_9_1: contains the patches applied when vllm version is 0.9.1.
# - patch_0_9_2: contains the patches applied when vllm version is 0.9.2.
# - patch_main: contains the patches applied when vllm version is main branch.
# - patch_common: contains the patches applied in both 0.9.1 and main branch.
# - patch_common: contains the patches applied in both 0.9.2 and main branch.
#
# Once a new patch is added in vllm-ascend, please add the patch description into this file as well.
# ----------------------------------------------------------------------------------
@@ -105,32 +105,6 @@
# Future Plan:
# Revert it when the related pr is merged in vllm and vllm-ascend.
#
# ** File: worker/patch_common/patch_sampler.py **
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# 1. `vllm.v1.sample.sampler.Sampler.apply_top_k_top_p`
# Why:
# We need to use the patched `apply_top_k_top_p` in `sample`.
# The mainly reason to overwrite `apply_top_k_top_p` is
# to improve performance.
# How：
# Re-implementation the `apply_top_k_top_p` function by pytorch
# Related PR (if no, explain why):
# - https://github.com/vllm-project/vllm-ascend/pull/970
# Future Plan:
# Revert it when the ascend scatter performance improves.
#
# 2. `vllm.v1.sample.sampler.Sampler.apply_min_p`
# Why:
# We need to use the patched `apply_min_p` in `sample`.
# The mainly reason to overwrite `apply_min_p` is
# to improve performance.
# How：
# Re-implementation the `apply_min_p` function by pytorch
# Related PR (if no, explain why):
# - https://github.com/vllm-project/vllm-ascend/pull/970
# Future Plan:
# Revert it when the ascend indexput performance improves.
#
# ** File: worker/patch_common/patch_distributed.py **
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# 1. `vllm.distributed.parallel_state.GroupCoordinator`
@@ -154,4 +128,4 @@
# Related PR (if no, explain why):
# This is the problem in vllm-ascend
# Future Plan:
# Remove this patch once pytorch 2.7.0 is supported for vllm ascend.
# Remove this patch once pytorch 2.7.0 is supported for vllm ascend.