xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-26 13:56:32 +08:00
Code Issues Packages Projects Releases Wiki Activity

[hardware][rocm] allow rocm to override default env var (#7926)

Browse Source
This commit is contained in:
youkaichao 2024-08-27 19:50:06 -07:00 committed by GitHub
parent fab5f53e2d
commit bc6e42a9b1
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 14 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/core/scheduler.py
View File

@ -1088,8 +1088,9 @@ class Scheduler:
)
def _allow_async_output_proc(self, seq_group: SequenceGroup) -> bool:
no_beam_search = (seq_group.sampling_params.best_of == 1
and not seq_group.sampling_params.use_beam_search)
no_beam_search = seq_group.sampling_params is None or (
seq_group.sampling_params.best_of == 1
and not seq_group.sampling_params.use_beam_search)
return no_beam_search

11
vllm/platforms/rocm.py
View File

@ -1,10 +1,21 @@
import os
from functools import lru_cache
from typing import Tuple
import torch
from vllm.logger import init_logger
from .interface import Platform, PlatformEnum
logger = init_logger(__name__)
if os.environ.get("VLLM_WORKER_MULTIPROC_METHOD", None) in ["fork", None]:
logger.warning("`fork` method is not supported by ROCm. "
"VLLM_WORKER_MULTIPROC_METHOD is overridden to"
" `spawn` instead.")
os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
class RocmPlatform(Platform):
_enum = PlatformEnum.ROCM