This commit is contained in:
+3
-17
@@ -1,5 +1,5 @@
|
||||
from .data import (EncoderDecoderLLMInputs, ExplicitEncoderDecoderPrompt,
|
||||
LLMInputs, PromptType, SingletonPrompt, TextPrompt,
|
||||
LLMInputs, PromptInputs, SingletonPromptInputs, TextPrompt,
|
||||
TokensPrompt, build_explicit_enc_dec_prompt,
|
||||
to_enc_dec_tuple_list, zip_enc_dec_prompts)
|
||||
from .registry import InputContext, InputRegistry
|
||||
@@ -16,8 +16,8 @@ See also:
|
||||
__all__ = [
|
||||
"TextPrompt",
|
||||
"TokensPrompt",
|
||||
"PromptType",
|
||||
"SingletonPrompt",
|
||||
"PromptInputs",
|
||||
"SingletonPromptInputs",
|
||||
"ExplicitEncoderDecoderPrompt",
|
||||
"LLMInputs",
|
||||
"EncoderDecoderLLMInputs",
|
||||
@@ -28,17 +28,3 @@ __all__ = [
|
||||
"InputContext",
|
||||
"InputRegistry",
|
||||
]
|
||||
|
||||
|
||||
def __getattr__(name: str):
|
||||
if name == "PromptInput":
|
||||
import warnings
|
||||
|
||||
msg = ("PromptInput has been renamed to PromptType. "
|
||||
"The original name will be removed in an upcoming version.")
|
||||
|
||||
warnings.warn(DeprecationWarning(msg), stacklevel=2)
|
||||
|
||||
return PromptType
|
||||
|
||||
raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
|
||||
|
||||
+19
-29
@@ -33,7 +33,7 @@ class TokensPrompt(TypedDict):
|
||||
"""
|
||||
|
||||
|
||||
SingletonPrompt = Union[str, TextPrompt, TokensPrompt]
|
||||
SingletonPromptInputs = Union[str, TextPrompt, TokensPrompt]
|
||||
"""
|
||||
Set of possible schemas for a single LLM input:
|
||||
|
||||
@@ -46,7 +46,7 @@ which may be utilized for encoder/decoder models when
|
||||
the user desires to express both the encoder & decoder
|
||||
prompts explicitly, i.e. :class:`ExplicitEncoderDecoderPrompt`
|
||||
|
||||
A prompt of type :class:`SingletonPrompt` may be employed
|
||||
A prompt of type :class:`SingletonPromptInputs` may be employed
|
||||
as (1) input to a decoder-only model, (2) input to
|
||||
the encoder of an encoder/decoder model, in the scenario
|
||||
where the decoder-prompt is not specified explicitly, or
|
||||
@@ -55,33 +55,33 @@ more than one prompt, i.e. :class:`ExplicitEncoderDecoderPrompt`
|
||||
"""
|
||||
|
||||
_T1_co = TypeVar("_T1_co",
|
||||
bound=SingletonPrompt,
|
||||
default=SingletonPrompt,
|
||||
bound=SingletonPromptInputs,
|
||||
default=SingletonPromptInputs,
|
||||
covariant=True)
|
||||
_T2_co = TypeVar("_T2_co",
|
||||
bound=SingletonPrompt,
|
||||
default=SingletonPrompt,
|
||||
bound=SingletonPromptInputs,
|
||||
default=SingletonPromptInputs,
|
||||
covariant=True)
|
||||
|
||||
|
||||
# TODO: Make fields ReadOnly once mypy supports it
|
||||
class ExplicitEncoderDecoderPrompt(TypedDict, Generic[_T1_co, _T2_co]):
|
||||
"""
|
||||
Represents an encoder/decoder model input prompt,
|
||||
comprising an explicit encoder prompt and a decoder prompt.
|
||||
"""Represents an encoder/decoder model input prompt,
|
||||
comprising an explicit encoder prompt and a
|
||||
decoder prompt.
|
||||
|
||||
The encoder and decoder prompts, respectively,
|
||||
may formatted according to any of the
|
||||
:class:`SingletonPrompt` schemas, and are not
|
||||
:class:`SingletonPromptInputs` schemas, and are not
|
||||
required to have the same schema.
|
||||
|
||||
Only the encoder prompt may have multi-modal data.
|
||||
|
||||
Note that an :class:`ExplicitEncoderDecoderPrompt` may not
|
||||
be used as an input to a decoder-only model,
|
||||
and that the :code:`encoder_prompt` and :code:`decoder_prompt`
|
||||
and that the `encoder_prompt` and `decoder_prompt`
|
||||
fields of this data structure themselves must be
|
||||
:class:`SingletonPrompt` instances.
|
||||
:class:`SingletonPromptInputs` instances.
|
||||
"""
|
||||
|
||||
encoder_prompt: _T1_co
|
||||
@@ -89,7 +89,7 @@ class ExplicitEncoderDecoderPrompt(TypedDict, Generic[_T1_co, _T2_co]):
|
||||
decoder_prompt: Optional[_T2_co]
|
||||
|
||||
|
||||
PromptType = Union[SingletonPrompt, ExplicitEncoderDecoderPrompt]
|
||||
PromptInputs = Union[SingletonPromptInputs, ExplicitEncoderDecoderPrompt]
|
||||
"""
|
||||
Set of possible schemas for an LLM input, including
|
||||
both decoder-only and encoder/decoder input types:
|
||||
@@ -140,8 +140,12 @@ class EncoderDecoderLLMInputs(LLMInputs):
|
||||
"""
|
||||
|
||||
|
||||
_T1 = TypeVar("_T1", bound=SingletonPrompt, default=SingletonPrompt)
|
||||
_T2 = TypeVar("_T2", bound=SingletonPrompt, default=SingletonPrompt)
|
||||
_T1 = TypeVar("_T1",
|
||||
bound=SingletonPromptInputs,
|
||||
default=SingletonPromptInputs)
|
||||
_T2 = TypeVar("_T2",
|
||||
bound=SingletonPromptInputs,
|
||||
default=SingletonPromptInputs)
|
||||
|
||||
|
||||
def build_explicit_enc_dec_prompt(
|
||||
@@ -172,17 +176,3 @@ def to_enc_dec_tuple_list(
|
||||
return [(enc_dec_prompt["encoder_prompt"],
|
||||
enc_dec_prompt["decoder_prompt"])
|
||||
for enc_dec_prompt in enc_dec_prompts]
|
||||
|
||||
|
||||
def __getattr__(name: str):
|
||||
if name == "PromptInput":
|
||||
import warnings
|
||||
|
||||
msg = ("PromptInput has been renamed to PromptType. "
|
||||
"The original name will be removed in an upcoming version.")
|
||||
|
||||
warnings.warn(DeprecationWarning(msg), stacklevel=2)
|
||||
|
||||
return PromptType
|
||||
|
||||
raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
|
||||
|
||||
+11
-11
@@ -5,7 +5,7 @@ from typing_extensions import TypeIs
|
||||
from vllm.utils import is_list_of
|
||||
|
||||
from .data import (EncoderDecoderLLMInputs, ExplicitEncoderDecoderPrompt,
|
||||
LLMInputs, PromptType, SingletonPrompt, TextPrompt,
|
||||
LLMInputs, PromptInputs, SingletonPromptInputs, TextPrompt,
|
||||
TokensPrompt)
|
||||
|
||||
|
||||
@@ -81,23 +81,23 @@ class ParsedTokensPrompt(TypedDict):
|
||||
|
||||
|
||||
def parse_singleton_prompt(
|
||||
prompt: SingletonPrompt,
|
||||
inputs: SingletonPromptInputs,
|
||||
) -> Union[ParsedStrPrompt, ParsedTextPrompt, ParsedTokensPrompt]:
|
||||
if isinstance(prompt, str):
|
||||
return ParsedStrPrompt(type="str", content=prompt)
|
||||
elif isinstance(prompt, dict):
|
||||
if "prompt_token_ids" in prompt:
|
||||
if isinstance(inputs, str):
|
||||
return ParsedStrPrompt(type="str", content=inputs)
|
||||
elif isinstance(inputs, dict):
|
||||
if "prompt_token_ids" in inputs:
|
||||
return ParsedTokensPrompt(type="tokens",
|
||||
content=prompt) # type: ignore
|
||||
elif "prompt" in prompt:
|
||||
return ParsedTextPrompt(type="text", content=prompt)
|
||||
content=inputs) # type: ignore
|
||||
elif "prompt" in inputs:
|
||||
return ParsedTextPrompt(type="text", content=inputs)
|
||||
|
||||
raise TypeError("inputs must be a string, TextPrompt, or TokensPrompt")
|
||||
|
||||
|
||||
def is_explicit_encoder_decoder_prompt(
|
||||
prompt: PromptType) -> TypeIs[ExplicitEncoderDecoderPrompt]:
|
||||
return isinstance(prompt, dict) and "encoder_prompt" in prompt
|
||||
inputs: PromptInputs) -> TypeIs[ExplicitEncoderDecoderPrompt]:
|
||||
return isinstance(inputs, dict) and "encoder_prompt" in inputs
|
||||
|
||||
|
||||
def is_valid_encoder_decoder_llm_inputs(
|
||||
|
||||
+43
-43
@@ -9,8 +9,8 @@ from vllm.lora.request import LoRARequest
|
||||
from vllm.prompt_adapter.request import PromptAdapterRequest
|
||||
from vllm.transformers_utils.tokenizer_group import BaseTokenizerGroup
|
||||
|
||||
from .data import (EncoderDecoderLLMInputs, LLMInputs, PromptType,
|
||||
SingletonPrompt)
|
||||
from .data import (EncoderDecoderLLMInputs, LLMInputs, PromptInputs,
|
||||
SingletonPromptInputs)
|
||||
from .parse import is_explicit_encoder_decoder_prompt, parse_singleton_prompt
|
||||
|
||||
if TYPE_CHECKING:
|
||||
@@ -206,7 +206,7 @@ class InputPreprocessor:
|
||||
|
||||
def _extract_prompt_components(
|
||||
self,
|
||||
prompt: SingletonPrompt,
|
||||
inputs: SingletonPromptInputs,
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
) -> PromptComponents:
|
||||
@@ -216,7 +216,7 @@ class InputPreprocessor:
|
||||
Arguments:
|
||||
|
||||
* request_id
|
||||
* prompt: single encoder or decoder input prompt
|
||||
* inputs: single encoder or decoder input prompt
|
||||
* lora_request: this is only valid for decoder prompts
|
||||
|
||||
Returns:
|
||||
@@ -226,24 +226,24 @@ class InputPreprocessor:
|
||||
* multi_modal_data
|
||||
'''
|
||||
|
||||
parsed = parse_singleton_prompt(prompt)
|
||||
parsed = parse_singleton_prompt(inputs)
|
||||
|
||||
if parsed["type"] == "str":
|
||||
prompt_text = parsed["content"]
|
||||
prompt = parsed["content"]
|
||||
prompt_token_ids = self._tokenize_prompt(
|
||||
prompt_text,
|
||||
prompt,
|
||||
request_id=request_id,
|
||||
lora_request=lora_request,
|
||||
)
|
||||
multi_modal_data = None
|
||||
elif parsed["type"] == "tokens":
|
||||
prompt_text = None
|
||||
prompt = None
|
||||
prompt_token_ids = parsed["content"]["prompt_token_ids"]
|
||||
multi_modal_data = parsed["content"].get("multi_modal_data")
|
||||
elif parsed["type"] == "text":
|
||||
prompt_text = parsed["content"]["prompt"]
|
||||
prompt = parsed["content"]["prompt"]
|
||||
prompt_token_ids = self._tokenize_prompt(
|
||||
prompt_text,
|
||||
prompt,
|
||||
request_id=request_id,
|
||||
lora_request=lora_request,
|
||||
)
|
||||
@@ -251,33 +251,33 @@ class InputPreprocessor:
|
||||
else:
|
||||
assert_never(parsed)
|
||||
|
||||
return prompt_text, prompt_token_ids, multi_modal_data
|
||||
return prompt, prompt_token_ids, multi_modal_data
|
||||
|
||||
async def _extract_prompt_components_async(
|
||||
self,
|
||||
prompt: SingletonPrompt,
|
||||
inputs: SingletonPromptInputs,
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
) -> PromptComponents:
|
||||
"""Async version of :meth:`_extract_prompt_components`."""
|
||||
parsed = parse_singleton_prompt(prompt)
|
||||
parsed = parse_singleton_prompt(inputs)
|
||||
|
||||
if parsed["type"] == "str":
|
||||
prompt_text = parsed["content"]
|
||||
prompt = parsed["content"]
|
||||
prompt_token_ids = await self._tokenize_prompt_async(
|
||||
prompt_text,
|
||||
prompt,
|
||||
request_id=request_id,
|
||||
lora_request=lora_request,
|
||||
)
|
||||
multi_modal_data = None
|
||||
elif parsed["type"] == "tokens":
|
||||
prompt_text = None
|
||||
prompt = None
|
||||
prompt_token_ids = parsed["content"]["prompt_token_ids"]
|
||||
multi_modal_data = parsed["content"].get("multi_modal_data")
|
||||
elif parsed["type"] == "text":
|
||||
prompt_text = parsed["content"]["prompt"]
|
||||
prompt = parsed["content"]["prompt"]
|
||||
prompt_token_ids = await self._tokenize_prompt_async(
|
||||
prompt_text,
|
||||
prompt,
|
||||
request_id=request_id,
|
||||
lora_request=lora_request,
|
||||
)
|
||||
@@ -285,7 +285,7 @@ class InputPreprocessor:
|
||||
else:
|
||||
assert_never(parsed)
|
||||
|
||||
return prompt_text, prompt_token_ids, multi_modal_data
|
||||
return prompt, prompt_token_ids, multi_modal_data
|
||||
|
||||
def _build_enc_dec_llm_inputs(
|
||||
self,
|
||||
@@ -311,7 +311,7 @@ class InputPreprocessor:
|
||||
|
||||
def _process_encoder_decoder_prompt(
|
||||
self,
|
||||
prompt: PromptType,
|
||||
inputs: PromptInputs,
|
||||
request_id: str,
|
||||
) -> EncoderDecoderLLMInputs:
|
||||
'''
|
||||
@@ -339,7 +339,7 @@ class InputPreprocessor:
|
||||
|
||||
Arguments:
|
||||
|
||||
* prompt: an input prompt
|
||||
* inputs: an input prompt
|
||||
* request_id
|
||||
|
||||
Returns:
|
||||
@@ -350,13 +350,13 @@ class InputPreprocessor:
|
||||
encoder_comps: PromptComponents
|
||||
decoder_comps: DecoderPromptComponents
|
||||
|
||||
if is_explicit_encoder_decoder_prompt(prompt):
|
||||
if is_explicit_encoder_decoder_prompt(inputs):
|
||||
encoder_comps = self._extract_prompt_components(
|
||||
prompt["encoder_prompt"],
|
||||
inputs["encoder_prompt"],
|
||||
request_id=request_id,
|
||||
)
|
||||
|
||||
if (decoder_input := prompt["decoder_prompt"]) is None:
|
||||
if (decoder_input := inputs["decoder_prompt"]) is None:
|
||||
decoder_comps = None, None, None
|
||||
else:
|
||||
decoder_comps = self._extract_prompt_components(
|
||||
@@ -365,7 +365,7 @@ class InputPreprocessor:
|
||||
)
|
||||
else:
|
||||
encoder_comps = self._extract_prompt_components(
|
||||
prompt,
|
||||
inputs,
|
||||
request_id=request_id,
|
||||
)
|
||||
|
||||
@@ -375,20 +375,20 @@ class InputPreprocessor:
|
||||
|
||||
async def _process_encoder_decoder_prompt_async(
|
||||
self,
|
||||
prompt: PromptType,
|
||||
inputs: PromptInputs,
|
||||
request_id: str,
|
||||
) -> EncoderDecoderLLMInputs:
|
||||
"""Async version of :meth:`_process_encoder_decoder_prompt`."""
|
||||
encoder_comps: PromptComponents
|
||||
decoder_comps: DecoderPromptComponents
|
||||
|
||||
if is_explicit_encoder_decoder_prompt(prompt):
|
||||
if is_explicit_encoder_decoder_prompt(inputs):
|
||||
encoder_task = self._extract_prompt_components_async(
|
||||
prompt["encoder_prompt"],
|
||||
inputs["encoder_prompt"],
|
||||
request_id=request_id,
|
||||
)
|
||||
|
||||
if (decoder_input := prompt["decoder_prompt"]) is None:
|
||||
if (decoder_input := inputs["decoder_prompt"]) is None:
|
||||
encoder_comps = await encoder_task
|
||||
decoder_comps = None, None, None
|
||||
else:
|
||||
@@ -401,7 +401,7 @@ class InputPreprocessor:
|
||||
encoder_task, decoder_task)
|
||||
else:
|
||||
encoder_comps = await self._extract_prompt_components_async(
|
||||
prompt,
|
||||
inputs,
|
||||
request_id=request_id,
|
||||
)
|
||||
|
||||
@@ -425,7 +425,7 @@ class InputPreprocessor:
|
||||
|
||||
def _process_decoder_only_prompt(
|
||||
self,
|
||||
prompt: SingletonPrompt,
|
||||
inputs: SingletonPromptInputs,
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
@@ -436,7 +436,7 @@ class InputPreprocessor:
|
||||
|
||||
Arguments:
|
||||
|
||||
* prompt: input prompt
|
||||
* inputs: input prompt
|
||||
* request_id
|
||||
* lora_request
|
||||
* prompt_adapter_request
|
||||
@@ -447,7 +447,7 @@ class InputPreprocessor:
|
||||
'''
|
||||
|
||||
prompt_comps = self._extract_prompt_components(
|
||||
prompt,
|
||||
inputs,
|
||||
request_id=request_id,
|
||||
lora_request=lora_request,
|
||||
)
|
||||
@@ -459,14 +459,14 @@ class InputPreprocessor:
|
||||
|
||||
async def _process_decoder_only_prompt_async(
|
||||
self,
|
||||
prompt: SingletonPrompt,
|
||||
inputs: SingletonPromptInputs,
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
) -> LLMInputs:
|
||||
"""Async version of :meth:`_process_decoder_only_prompt`."""
|
||||
prompt_comps = await self._extract_prompt_components_async(
|
||||
prompt,
|
||||
inputs,
|
||||
request_id=request_id,
|
||||
lora_request=lora_request,
|
||||
)
|
||||
@@ -478,7 +478,7 @@ class InputPreprocessor:
|
||||
|
||||
def preprocess(
|
||||
self,
|
||||
prompt: PromptType,
|
||||
inputs: PromptInputs,
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
@@ -488,17 +488,17 @@ class InputPreprocessor:
|
||||
# Encoder-decoder model requires special mapping of
|
||||
# input prompts to encoder & decoder
|
||||
return self._process_encoder_decoder_prompt(
|
||||
prompt,
|
||||
inputs,
|
||||
request_id=request_id,
|
||||
)
|
||||
|
||||
if is_explicit_encoder_decoder_prompt(prompt):
|
||||
if is_explicit_encoder_decoder_prompt(inputs):
|
||||
raise ValueError("Cannot pass encoder-decoder prompt "
|
||||
"to decoder-only models")
|
||||
|
||||
# Decoder-only operation
|
||||
return self._process_decoder_only_prompt(
|
||||
prompt,
|
||||
inputs,
|
||||
request_id=request_id,
|
||||
lora_request=lora_request,
|
||||
prompt_adapter_request=prompt_adapter_request,
|
||||
@@ -506,7 +506,7 @@ class InputPreprocessor:
|
||||
|
||||
async def preprocess_async(
|
||||
self,
|
||||
prompt: PromptType,
|
||||
inputs: PromptInputs,
|
||||
request_id: str,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
|
||||
@@ -516,17 +516,17 @@ class InputPreprocessor:
|
||||
# Encoder-decoder model requires special mapping of
|
||||
# input prompts to encoder & decoder
|
||||
return await self._process_encoder_decoder_prompt_async(
|
||||
prompt,
|
||||
inputs,
|
||||
request_id=request_id,
|
||||
)
|
||||
|
||||
if is_explicit_encoder_decoder_prompt(prompt):
|
||||
if is_explicit_encoder_decoder_prompt(inputs):
|
||||
raise ValueError("Cannot pass encoder-decoder prompt "
|
||||
"to decoder-only models")
|
||||
|
||||
# Decoder-only operation
|
||||
return await self._process_decoder_only_prompt_async(
|
||||
prompt,
|
||||
inputs,
|
||||
request_id=request_id,
|
||||
lora_request=lora_request,
|
||||
prompt_adapter_request=prompt_adapter_request,
|
||||
|
||||
Reference in New Issue
Block a user