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

9
vllm/model_executor/models/gemma2.py
View File

@@ -312,7 +312,8 @@ class Gemma2Model(nn.Module):
hidden_states, _ = self.norm(hidden_states, residual)
return hidden_states
def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
def load_weights(self, weights: Iterable[Tuple[str,
torch.Tensor]]) -> Set[str]:
stacked_params_mapping = [
# (param_name, shard_name, shard_id)
("qkv_proj", "q_proj", "q"),
@@ -354,6 +355,7 @@ class Gemma2Model(nn.Module):
logger.warning(
"Some weights are not initialized from checkpoints: %s",
unloaded_params)
return loaded_params
class Gemma2ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
@@ -451,13 +453,14 @@ class Gemma2ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
next_tokens = self.sampler(logits, sampling_metadata)
return next_tokens
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,
skip_prefixes=(["lm_head."]
if self.config.tie_word_embeddings else None),
)
loader.load_weights(weights)
return loader.load_weights(weights)
class Gemma2EmbeddingModel(nn.Module, SupportsPP):