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

upgrade vLLM to 0.12.0 tag (#4647)

Browse Source 
Upgrade vLLM to v0.12.0 tag

- vLLM version: 86e178f7c4d8c3b0eaf3c8e3f810a83f63b90e24
- vLLM main:
86e178f7c4

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-12-03 23:43:05 +08:00
committed by GitHub
parent 26e8e58cea
commit 3f4c0ea0a0
22 changed files with 97 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
vllm_ascend/platform.py
View File

@@ -159,7 +159,8 @@ class NPUPlatform(Platform):
compilation_config.splitting_ops = []
compilation_config.cudagraph_num_of_warmups = 1
compilation_config.pass_config.enable_fusion = False
compilation_config.pass_config.fuse_norm_quant = False
compilation_config.pass_config.fuse_act_quant = False
if compilation_config.mode not in [
CompilationMode.NONE, CompilationMode.VLLM_COMPILE
@@ -194,7 +195,7 @@ class NPUPlatform(Platform):
# to ascend ops && hardwares. We update these sizes here to improve
# default performance.
update_default_aclgraph_sizes(vllm_config)
# TODO delete graph size update here when compilation_config.pass_config.enable_sequence_parallelism
# TODO delete graph size update here when compilation_config.pass_config.enable_sp
# is supported by vllm-ascend.
if vllm_config.parallel_config.tensor_parallel_size > 1 and not vllm_config.model_config.enforce_eager and \
enable_sp(vllm_config):