xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-09 23:06:10 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix][Model] Fix inference for Hunyuan dense models (#25354)

Browse Source 
Signed-off-by: anion <1005128408@qq.com>
Signed-off-by: Anion <123177548+Anionex@users.noreply.github.com>
This commit is contained in:
Anion 2025-09-30 22:38:07 +08:00 committed by GitHub
parent 099aaee536
commit f4db5e6de1
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 60 additions and 48 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

108
vllm/model_executor/models/hunyuan_v1.py
View File

@ -888,7 +888,7 @@ class HunYuanModel(nn.Module):
return loaded_params
class HunYuanV1Base(nn.Module, SupportsLoRA, SupportsPP, MixtureOfExperts):
class HunyuanV1ModelBase(nn.Module, SupportsLoRA, SupportsPP):
packed_modules_mapping = {
"qkv_proj": [
"q_proj",
@ -930,6 +930,56 @@ class HunYuanV1Base(nn.Module, SupportsLoRA, SupportsPP, MixtureOfExperts):
else:
self.lm_head = PPMissingLayer()
def forward(
self,
input_ids: torch.Tensor,
positions: torch.Tensor,
intermediate_tensors: Optional[IntermediateTensors] = None,
inputs_embeds: Optional[torch.Tensor] = None,
) -> Union[torch.Tensor, IntermediateTensors]:
model_output = self.model(input_ids, positions, intermediate_tensors,
inputs_embeds)
return model_output
def compute_logits(
self,
hidden_states: torch.Tensor,
) -> Optional[torch.Tensor]:
logits = self.logits_processor(self.lm_head, hidden_states)
return logits
def make_empty_intermediate_tensors(
self, batch_size: int, dtype: torch.dtype,
device: torch.device) -> IntermediateTensors:
return IntermediateTensors({
"hidden_states":
torch.zeros((batch_size, self.config.hidden_size),
dtype=dtype,
device=device),
"residual":
torch.zeros((batch_size, self.config.hidden_size),
dtype=dtype,
device=device),
})
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),
)
return loader.load_weights(weights)
def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
return self.model.get_input_embeddings(input_ids)
class HunYuanMoEV1Base(HunyuanV1ModelBase, MixtureOfExperts):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__(vllm_config=vllm_config, prefix=prefix)
# Set MoE hyperparameters
self.expert_weights = []
self.num_expert_groups = 1
@ -988,57 +1038,19 @@ class HunYuanV1Base(nn.Module, SupportsLoRA, SupportsPP, MixtureOfExperts):
moe.n_redundant_experts = self.num_redundant_experts
moe.experts.update_expert_map()
def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
return self.model.get_input_embeddings(input_ids)
def forward(
self,
input_ids: torch.Tensor,
positions: torch.Tensor,
intermediate_tensors: Optional[IntermediateTensors] = None,
inputs_embeds: Optional[torch.Tensor] = None,
) -> Union[torch.Tensor, IntermediateTensors]:
model_output = self.model(input_ids, positions, intermediate_tensors,
inputs_embeds)
return model_output
def compute_logits(
self,
hidden_states: torch.Tensor,
) -> Optional[torch.Tensor]:
logits = self.logits_processor(self.lm_head, hidden_states)
return logits
def make_empty_intermediate_tensors(
self, batch_size: int, dtype: torch.dtype,
device: torch.device) -> IntermediateTensors:
return IntermediateTensors({
"hidden_states":
torch.zeros((batch_size, self.config.hidden_size),
dtype=dtype,
device=device),
"residual":
torch.zeros((batch_size, self.config.hidden_size),
dtype=dtype,
device=device),
})
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),
)
return loader.load_weights(weights)
def get_expert_mapping(self) -> list[tuple[str, str, int, str]]:
return self.model.get_expert_mapping()
class HunYuanDenseV1ForCausalLM(HunYuanV1Base):
class HunYuanDenseV1Base(HunyuanV1ModelBase):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__(vllm_config=vllm_config, prefix=prefix)
class HunYuanDenseV1ForCausalLM(HunYuanDenseV1Base):
pass
class HunYuanMoEV1ForCausalLM(HunYuanV1Base):
pass
class HunYuanMoEV1ForCausalLM(HunYuanMoEV1Base):
pass