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

[V1] VLM - Run the mm_mapper preprocessor in the frontend process (#10640)

Browse Source 
Signed-off-by: Roger Wang <ywang@roblox.com>
Co-authored-by: Michael Goin <michael@neuralmagic.com>
Co-authored-by: Roger Wang <ywang@roblox.com>
This commit is contained in:
Alexander Matveev
2024-12-03 05:33:10 -05:00
committed by GitHub
parent f6084f6324
commit 3bc94cab69
7 changed files with 47 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
vllm/v1/request.py
View File

@@ -45,9 +45,6 @@ class Request:
self._all_token_ids: List[int] = self.prompt_token_ids.copy()
self.num_computed_tokens = 0
# Raw multimodal data before the mm input mapper (e.g., PIL images).
self.mm_data = self.inputs.multi_modal_data
self.mm_processor_kwargs = self.inputs.mm_processor_kwargs
mm_positions = self.inputs.multi_modal_placeholders
if mm_positions:
# FIXME(woosuk): Support other modalities.
@@ -55,7 +52,10 @@ class Request:
else:
self.mm_positions = []
# Output of the mm input mapper (e.g., image tensors).
self.mm_inputs: List[MultiModalKwargs] = []
if self.inputs.multi_modal_inputs:
self.mm_inputs = self.inputs.multi_modal_inputs
else:
self.mm_inputs: List[MultiModalKwargs] = []
@classmethod
def from_engine_core_request(cls, request: EngineCoreRequest) -> "Request":
@@ -64,9 +64,10 @@ class Request:
inputs=token_inputs(
prompt_token_ids=request.prompt_token_ids,
prompt=request.prompt,
multi_modal_data=request.mm_data,
multi_modal_data=None,
multi_modal_inputs=request.mm_inputs,
multi_modal_placeholders=request.mm_placeholders,
mm_processor_kwargs=request.mm_processor_kwargs,
mm_processor_kwargs=None,
),
sampling_params=request.sampling_params,
eos_token_id=request.eos_token_id,
@@ -110,7 +111,7 @@ class Request:
return RequestStatus.get_finished_reason(self.status)
def has_encoder_inputs(self) -> bool:
return len(self.mm_data) > 0
return len(self.mm_inputs) > 0
@property
def num_encoder_inputs(self) -> int: