biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[MISC] Introduce pipeline parallelism partition strategies (#6920)

Browse Source 
Co-authored-by: youkaichao <youkaichao@126.com>
This commit is contained in:
Cody Yu
2024-07-31 12:02:17 -07:00
committed by GitHub
parent 2ee8d3ba55
commit bd70013407
3 changed files with 66 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/envs.py
View File

@@ -28,6 +28,7 @@ if TYPE_CHECKING:
VLLM_LOGGING_CONFIG_PATH: Optional[str] = None
VLLM_TRACE_FUNCTION: int = 0
VLLM_ATTENTION_BACKEND: Optional[str] = None
VLLM_PP_LAYER_PARTITION: Optional[str] = None
VLLM_CPU_KVCACHE_SPACE: int = 0
VLLM_CPU_OMP_THREADS_BIND: str = ""
VLLM_OPENVINO_KVCACHE_SPACE: int = 0
@@ -242,6 +243,10 @@ environment_variables: Dict[str, Callable[[], Any]] = {
"VLLM_ATTENTION_BACKEND":
lambda: os.getenv("VLLM_ATTENTION_BACKEND", None),
# Pipeline stage partition strategy
"VLLM_PP_LAYER_PARTITION":
lambda: os.getenv("VLLM_PP_LAYER_PARTITION", None),
# (CPU backend only) CPU key-value cache space.
# default is 4GB
"VLLM_CPU_KVCACHE_SPACE":