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

[Misc] Drop Prefetch MLP Env (#7357)

Browse Source 
### What this PR does / why we need it?
remove deprecated environment variables related to MLP prefetching

### Does this PR introduce _any_ user-facing change?
yes, the deprecated env vars can not be used then.

- vLLM version: v0.17.0
- vLLM main:
4034c3d32e

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2026-03-19 14:27:27 +08:00
committed by GitHub
parent ce239db4fb
commit 8e0ebb470a
4 changed files with 10 additions and 65 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
vllm_ascend/envs.py
View File

@@ -77,16 +77,6 @@ env_variables: dict[str, Callable[[], Any]] = {
# For a detailed introduction to the parameters and the differences and applicable scenarios
# between this feature and FLASHCOMM1, please refer to the feature guide in the documentation.
"VLLM_ASCEND_FLASHCOMM2_PARALLEL_SIZE": lambda: int(os.getenv("VLLM_ASCEND_FLASHCOMM2_PARALLEL_SIZE", 0)),
# Whether to enable MLP weight prefetch, only used in small concurrency.
"VLLM_ASCEND_ENABLE_PREFETCH_MLP": lambda: bool(int(os.getenv("VLLM_ASCEND_ENABLE_PREFETCH_MLP", "0"))),
# buffer size for gate up prefetch
"VLLM_ASCEND_MLP_GATE_UP_PREFETCH_SIZE": lambda: int(
os.getenv("VLLM_ASCEND_MLP_GATE_UP_PREFETCH_SIZE", 18 * 1024 * 1024)
),
# buffer size for down proj prefetch
"VLLM_ASCEND_MLP_DOWN_PREFETCH_SIZE": lambda: int(
os.getenv("VLLM_ASCEND_MLP_DOWN_PREFETCH_SIZE", 18 * 1024 * 1024)
),
# Whether to enable msMonitor tool to monitor the performance of vllm-ascend.
"MSMONITOR_USE_DAEMON": lambda: bool(int(os.getenv("MSMONITOR_USE_DAEMON", "0"))),
# Whether to enable MLAPO optimization for DeepSeek W8A8 series models.