[Core] Remove legacy input mapper/processor from V0 (#15686)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-04-28 15:38:48 +08:00
committed by GitHub
parent 7fcc4223dc
commit aec9674dbe
21 changed files with 142 additions and 1337 deletions
+17 -30
View File
@@ -14,9 +14,9 @@ from typing import Any, Callable, Optional, Union
import msgspec
import torch
from vllm.inputs import SingletonInputs, SingletonInputsAdapter
from vllm.inputs import SingletonInputs
from vllm.lora.request import LoRARequest
from vllm.multimodal import MultiModalDataDict, MultiModalPlaceholderDict
from vllm.multimodal import MultiModalKwargs, MultiModalPlaceholderDict
from vllm.pooling_params import PoolingParams
from vllm.prompt_adapter.request import PromptAdapterRequest
from vllm.sampling_params import RequestOutputKind, SamplingParams
@@ -419,7 +419,7 @@ class Sequence:
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
) -> None:
self.seq_id = seq_id
self.inputs = SingletonInputsAdapter(inputs)
self.inputs = inputs
self.block_size = block_size
self.eos_token_id = eos_token_id
self.lora_request = lora_request
@@ -448,31 +448,29 @@ class Sequence:
@property
def prompt(self) -> Optional[str]:
return self.inputs.prompt
return self.inputs.get("prompt")
@property
def prompt_token_ids(self) -> list[int]:
return self.inputs.prompt_token_ids
@property
def prompt_embeds(self) -> Optional[torch.Tensor]:
return self.inputs.prompt_embeds
return self.inputs["prompt_token_ids"]
@property
def token_type_ids(self) -> list[int]:
return self.inputs.token_type_ids
return self.inputs.get("token_type_ids", [])
@property
def multi_modal_data(self) -> "MultiModalDataDict":
return self.inputs.multi_modal_data
def multi_modal_data(self) -> MultiModalKwargs:
if self.inputs["type"] == "multimodal":
return self.inputs["mm_kwargs"]
return MultiModalKwargs({})
@property
def multi_modal_placeholders(self) -> MultiModalPlaceholderDict:
return self.inputs.multi_modal_placeholders
if self.inputs["type"] == "multimodal":
return self.inputs["mm_placeholders"]
@property
def mm_processor_kwargs(self) -> dict[str, Any]:
return self.inputs.mm_processor_kwargs
return {}
@property
def lora_int_id(self) -> int:
@@ -723,12 +721,12 @@ class SequenceGroup:
return self.first_seq.token_type_ids
@property
def multi_modal_data(self) -> MultiModalDataDict:
def multi_modal_data(self) -> MultiModalKwargs:
if self.first_seq.multi_modal_data:
return self.first_seq.multi_modal_data
elif self.encoder_seq is not None:
return self.encoder_seq.multi_modal_data
return {}
return MultiModalKwargs({})
@property
def multi_modal_placeholders(self) -> MultiModalPlaceholderDict:
@@ -738,14 +736,6 @@ class SequenceGroup:
return self.encoder_seq.multi_modal_placeholders
return {}
@property
def mm_processor_kwargs(self) -> dict[str, Any]:
if self.first_seq.multi_modal_data:
return self.first_seq.mm_processor_kwargs
elif self.encoder_seq is not None:
return self.encoder_seq.mm_processor_kwargs
return {}
@property
def lora_int_id(self) -> int:
return self.lora_request.lora_int_id if self.lora_request else 0
@@ -969,12 +959,9 @@ class SequenceGroupMetadata(
computed_block_nums: Optional[list[int]] = None
state: Optional[SequenceGroupState] = msgspec.field(
default_factory=lambda: SequenceGroupState())
# "MultiModalDataDict" types. We have to use Any due to msgspec
# doesn't allow to have union of 2 different dicts.
token_type_ids: Optional[list[int]] = None
multi_modal_data: Optional[Any] = None
multi_modal_data: Optional[MultiModalKwargs] = None
multi_modal_placeholders: Optional[MultiModalPlaceholderDict] = None
mm_processor_kwargs: Optional[dict[str, Any]] = None
encoder_seq_data: Optional[SequenceData] = None
cross_block_table: Optional[list[int]] = None
prompt_adapter_request: Optional[PromptAdapterRequest] = None