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

support FULL graph mode for GQA (#3970)

Browse Source 
### What this PR does / why we need it?
The current library only supports the FullDecodeOnly graph mode, which
enables full graph execution during the decode. This PR extends support
to allow full graph execution in both the prefill and decode, referred
to as FULL graph mode.

- vLLM version: v0.11.0
- vLLM main:
2918c1b49c

Signed-off-by: wangxiaoxin-sherie <wangxiaoxin7@huawei.com>
Co-authored-by: wangxiaoxin-sherie <wangxiaoxin7@huawei.com>
This commit is contained in:
XiaoxinWang
2025-11-17 10:50:35 +08:00
committed by GitHub
parent c334114f69
commit e38ef2c434
11 changed files with 328 additions and 296 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm_ascend/platform.py
View File

@@ -233,7 +233,8 @@ class NPUPlatform(Platform):
"vllm.mla_forward"
])
update_aclgraph_sizes(vllm_config)
elif compilation_config.cudagraph_mode == CUDAGraphMode.FULL_DECODE_ONLY:
elif compilation_config.cudagraph_mode == CUDAGraphMode.FULL_DECODE_ONLY or\
compilation_config.cudagraph_mode == CUDAGraphMode.FULL:
logger.info(
"FULL_DECODE_ONLY compilation enabled on NPU. use_inductor not supported - "
"using only ACL Graph mode")
@@ -270,7 +271,8 @@ class NPUPlatform(Platform):
compilation_config.use_inductor = False
compilation_config.splitting_ops.extend(["vllm::mla_forward"])
update_aclgraph_sizes(vllm_config)
elif compilation_config.cudagraph_mode == CUDAGraphMode.FULL_DECODE_ONLY:
elif compilation_config.cudagraph_mode == CUDAGraphMode.FULL_DECODE_ONLY or\
compilation_config.cudagraph_mode == CUDAGraphMode.FULL:
logger.info(
"FULL_DECODE_ONLY compilation enabled on NPU. use_inductor not supported - "
"using only ACL Graph mode")