[Core] Rename input data types (#8688)
This commit is contained in:
+18
-18
@@ -10,7 +10,7 @@ from vllm.prompt_adapter.request import PromptAdapterRequest
|
||||
from vllm.transformers_utils.tokenizer_group import BaseTokenizerGroup
|
||||
from vllm.utils import print_warning_once
|
||||
|
||||
from .data import (EncoderDecoderLLMInputs, LLMInputs, PromptType,
|
||||
from .data import (DecoderOnlyInputs, EncoderDecoderInputs, PromptType,
|
||||
SingletonPrompt)
|
||||
from .parse import is_explicit_encoder_decoder_prompt, parse_singleton_prompt
|
||||
|
||||
@@ -306,7 +306,7 @@ class InputPreprocessor:
|
||||
encoder_comps: PromptComponents,
|
||||
decoder_comps: DecoderPromptComponents,
|
||||
mm_processor_kwargs: Dict[str, Any],
|
||||
) -> EncoderDecoderLLMInputs:
|
||||
) -> EncoderDecoderInputs:
|
||||
encoder_prompt, encoder_prompt_ids, encoder_mm_data, _ = encoder_comps
|
||||
decoder_prompt, decoder_prompt_ids, decoder_mm_data, _ = decoder_comps
|
||||
|
||||
@@ -324,7 +324,7 @@ class InputPreprocessor:
|
||||
decoder_prompt_ids,
|
||||
force_bos=(encoder_mm_data is None and decoder_mm_data is None)))
|
||||
|
||||
return EncoderDecoderLLMInputs(
|
||||
return EncoderDecoderInputs(
|
||||
prompt_token_ids=decoder_prompt_ids,
|
||||
prompt=decoder_prompt,
|
||||
multi_modal_data=decoder_mm_data,
|
||||
@@ -338,11 +338,11 @@ class InputPreprocessor:
|
||||
self,
|
||||
prompt: PromptType,
|
||||
request_id: str,
|
||||
) -> EncoderDecoderLLMInputs:
|
||||
) -> EncoderDecoderInputs:
|
||||
'''
|
||||
For encoder/decoder models only:
|
||||
Process an input prompt into an
|
||||
:class:`EncoderDecoderLLMInputs` instance.
|
||||
:class:`EncoderDecoderInputs` instance.
|
||||
|
||||
There are two types of input prompts:
|
||||
singleton prompts which carry only the
|
||||
@@ -369,7 +369,7 @@ class InputPreprocessor:
|
||||
|
||||
Returns:
|
||||
|
||||
* :class:`EncoderDecoderLLMInputs` instance
|
||||
* :class:`EncoderDecoderInputs` instance
|
||||
'''
|
||||
|
||||
encoder_comps: PromptComponents
|
||||
@@ -411,7 +411,7 @@ class InputPreprocessor:
|
||||
self,
|
||||
prompt: PromptType,
|
||||
request_id: str,
|
||||
) -> EncoderDecoderLLMInputs:
|
||||
) -> EncoderDecoderInputs:
|
||||
"""Async version of :meth:`_process_encoder_decoder_prompt`."""
|
||||
encoder_comps: PromptComponents
|
||||
decoder_comps: DecoderPromptComponents
|
||||
@@ -455,17 +455,17 @@ class InputPreprocessor:
|
||||
self,
|
||||
prompt_comps: PromptComponents,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest],
|
||||
) -> LLMInputs:
|
||||
) -> DecoderOnlyInputs:
|
||||
(prompt, prompt_token_ids, multi_modal_data,
|
||||
mm_processor_kwargs) = prompt_comps
|
||||
|
||||
prompt_token_ids = self._apply_prompt_adapter(
|
||||
prompt_token_ids, prompt_adapter_request=prompt_adapter_request)
|
||||
|
||||
return LLMInputs(prompt_token_ids=prompt_token_ids,
|
||||
prompt=prompt,
|
||||
multi_modal_data=multi_modal_data,
|
||||
mm_processor_kwargs=mm_processor_kwargs)
|
||||
return DecoderOnlyInputs(prompt_token_ids=prompt_token_ids,
|
||||
prompt=prompt,
|
||||
multi_modal_data=multi_modal_data,
|
||||
mm_processor_kwargs=mm_processor_kwargs)
|
||||
|
||||
def _process_decoder_only_prompt(
|
||||
self,
|
||||
@@ -473,10 +473,10 @@ class InputPreprocessor:
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
) -> LLMInputs:
|
||||
) -> DecoderOnlyInputs:
|
||||
'''
|
||||
For decoder-only models:
|
||||
Process an input prompt into an :class:`LLMInputs` instance.
|
||||
Process an input prompt into an :class:`DecoderOnlyInputs` instance.
|
||||
|
||||
Arguments:
|
||||
|
||||
@@ -487,7 +487,7 @@ class InputPreprocessor:
|
||||
|
||||
Returns:
|
||||
|
||||
* :class:`LLMInputs` instance
|
||||
* :class:`DecoderOnlyInputs` instance
|
||||
'''
|
||||
|
||||
prompt_comps = self._extract_prompt_components(
|
||||
@@ -507,7 +507,7 @@ class InputPreprocessor:
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
) -> LLMInputs:
|
||||
) -> DecoderOnlyInputs:
|
||||
"""Async version of :meth:`_process_decoder_only_prompt`."""
|
||||
prompt_comps = await self._extract_prompt_components_async(
|
||||
prompt,
|
||||
@@ -526,7 +526,7 @@ class InputPreprocessor:
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
) -> Union[LLMInputs, EncoderDecoderLLMInputs]:
|
||||
) -> Union[DecoderOnlyInputs, EncoderDecoderInputs]:
|
||||
"""Preprocess the input prompt."""
|
||||
if self.is_encoder_decoder_model():
|
||||
# Encoder-decoder model requires special mapping of
|
||||
@@ -554,7 +554,7 @@ class InputPreprocessor:
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
) -> Union[LLMInputs, EncoderDecoderLLMInputs]:
|
||||
) -> Union[DecoderOnlyInputs, EncoderDecoderInputs]:
|
||||
"""Async version of :meth:`preprocess`."""
|
||||
if self.is_encoder_decoder_model():
|
||||
# Encoder-decoder model requires special mapping of
|
||||
|
||||
Reference in New Issue
Block a user