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

[Misc] Add uninitialized params tracking for AutoWeightsLoader (#10327)

Browse Source 
Signed-off-by: Isotr0py <2037008807@qq.com>
This commit is contained in:
Isotr0py
2024-11-18 09:07:46 +08:00
committed by GitHub
parent d1557e66d3
commit c4e464333e
74 changed files with 454 additions and 185 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/model_executor/models/qwen2_rm.py
View File

@@ -3,7 +3,7 @@
# Copyright 2024 The Qwen team.
# Copyright 2023 The vLLM team.
"""Inference-only Qwen2-RM model compatible with HuggingFace weights."""
from typing import Iterable, List, Optional, Tuple, Union
from typing import Iterable, List, Optional, Set, Tuple, Union
import torch
from torch import nn
@@ -110,7 +110,8 @@ class Qwen2ForRewardModel(nn.Module, SupportsLoRA, SupportsPP):
) -> Optional[PoolerOutput]:
return self._pooler(hidden_states, pooling_metadata)
def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
def load_weights(self, weights: Iterable[Tuple[str,
torch.Tensor]]) -> Set[str]:
loader = AutoWeightsLoader(self,
ignore_unexpected_prefixes=["lm_head."])
loader.load_weights(weights)
return loader.load_weights(weights)