frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Optimization] Use a cheaper cache key in get_model_architecture (#25682)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-09-26 05:54:20 +08:00
committed by GitHub
parent 3d54bdcb73
commit 89fa54e6f7
1 changed files with 9 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
vllm/model_executor/model_loader/utils.py
View File

@ -165,7 +165,7 @@ def device_loading_context(module: torch.nn.Module,
# New parameters or parameters already on target device are untouched
_MODEL_ARCH_BY_HASH = dict[str, tuple[type[nn.Module], str]]()
_MODEL_ARCH_BY_HASH = dict[int, tuple[type[nn.Module], str]]()
"""Caches the outputs of `_get_model_architecture`."""
@ -215,7 +215,14 @@ def _get_model_architecture(
def get_model_architecture(
model_config: ModelConfig) -> tuple[type[nn.Module], str]:
key = model_config.compute_hash()
key = hash((
model_config.model,
model_config.convert_type,
model_config.runner_type,
model_config.trust_remote_code,
model_config.model_impl,
tuple(getattr(model_config.hf_config, "architectures", [])),
))
if key in _MODEL_ARCH_BY_HASH:
return _MODEL_ARCH_BY_HASH[key]