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

[V1] Refactor model executable interface for all text-only language models (#10374)

Browse Source 
Signed-off-by: Roger Wang <ywang@roblox.com>
This commit is contained in:
Roger Wang
2024-11-16 21:18:46 -08:00
committed by GitHub
parent 4fd9375028
commit 643ecf7b11
43 changed files with 483 additions and 90 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
vllm/model_executor/models/starcoder2.py
View File

@@ -221,6 +221,9 @@ class Starcoder2Model(nn.Module):
make_empty_intermediate_tensors_factory(["hidden_states"],
config.hidden_size))
def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
return self.embed_tokens(input_ids)
def forward(
self,
input_ids: torch.Tensor,
@@ -228,9 +231,13 @@ class Starcoder2Model(nn.Module):
kv_caches: List[torch.Tensor],
attn_metadata: AttentionMetadata,
intermediate_tensors: Optional[IntermediateTensors],
inputs_embeds: Optional[torch.Tensor] = None,
) -> Union[torch.Tensor, IntermediateTensors]:
if get_pp_group().is_first_rank:
hidden_states = self.embed_tokens(input_ids)
if inputs_embeds is not None:
hidden_states = inputs_embeds
else:
hidden_states = self.get_input_embeddings(input_ids)
else:
assert intermediate_tensors is not None
hidden_states = intermediate_tensors["hidden_states"]
@@ -273,6 +280,9 @@ class Starcoder2ForCausalLM(nn.Module, SupportsPP):
self.make_empty_intermediate_tensors = (
self.model.make_empty_intermediate_tensors)
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,
@@ -280,9 +290,11 @@ class Starcoder2ForCausalLM(nn.Module, SupportsPP):
kv_caches: List[torch.Tensor],
attn_metadata: AttentionMetadata,
intermediate_tensors: Optional[IntermediateTensors] = None,
inputs_embeds: Optional[torch.Tensor] = None,
) -> Union[torch.Tensor, IntermediateTensors]:
hidden_states = self.model(input_ids, positions, kv_caches,
attn_metadata, intermediate_tensors)
attn_metadata, intermediate_tensors,
inputs_embeds)
return hidden_states
def compute_logits(