[Core] [Bugfix] Add Input Embeddings (#15428)

Signed-off-by: Andrew Sansom <andrew@protopia.ai>
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Co-authored-by: 临景 <linjing.yx@alibaba-inc.com>
Co-authored-by: Bryce1010 <bryceyx@gmail.com>
Co-authored-by: Nan2018 <nan@protopia.ai>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
Co-authored-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Andrew Sansom
2025-05-02 01:06:39 -07:00
committed by GitHub
co-authored by 临景 Bryce1010 Nan2018 Cyrus Leung DarkLight1337
parent 9e2de9b9e9
commit cc2a77d7f1
22 changed files with 691 additions and 113 deletions
+48 -8
View File
@@ -6,8 +6,9 @@ from typing_extensions import TypeIs
from vllm.utils import is_list_of
from .data import (ExplicitEncoderDecoderPrompt, ProcessorInputs, PromptType,
SingletonInputs, SingletonPrompt, TextPrompt, TokensPrompt)
from .data import (EmbedsInputs, EmbedsPrompt, ExplicitEncoderDecoderPrompt,
ProcessorInputs, PromptType, SingletonInputs,
SingletonPrompt, TextPrompt, TokensPrompt)
class ParsedText(TypedDict):
@@ -84,30 +85,69 @@ class ParsedTokensPrompt(TypedDict):
content: TokensPrompt
class ParsedEmbedsPrompt(TypedDict):
type: Literal['embeds']
content: EmbedsPrompt
@overload
def parse_singleton_prompt(prompt: str) -> ParsedStrPrompt:
...
@overload
def parse_singleton_prompt(prompt: TextPrompt) -> ParsedTextPrompt:
...
@overload
def parse_singleton_prompt(prompt: TokensPrompt) -> ParsedTokensPrompt:
...
@overload
def parse_singleton_prompt(prompt: EmbedsPrompt) -> ParsedEmbedsPrompt:
...
def parse_singleton_prompt(
prompt: SingletonPrompt,
) -> Union[ParsedStrPrompt, ParsedTextPrompt, ParsedTokensPrompt]:
) -> Union[ParsedStrPrompt, ParsedTextPrompt, ParsedTokensPrompt,
ParsedEmbedsPrompt]:
if isinstance(prompt, str):
return ParsedStrPrompt(type="str", content=prompt)
elif isinstance(prompt, dict):
if "prompt_token_ids" in prompt:
return ParsedTokensPrompt(type="tokens",
content=prompt) # type: ignore
# Type ignores are because mypy does not correctly infer the TypedDicts
# Pyright does succeed.
if "prompt_embeds" in prompt:
return ParsedEmbedsPrompt(
type="embeds", content=prompt) # type: ignore[typeddict-item]
elif "prompt_token_ids" in prompt:
return ParsedTokensPrompt(
type="tokens", content=prompt) # type: ignore[typeddict-item]
elif "prompt" in prompt:
return ParsedTextPrompt(type="text", content=prompt)
raise TypeError("inputs must be a string, TextPrompt, or TokensPrompt")
raise TypeError(
"inputs must be a string, TextPrompt, TokensPrompt, or EmbedsPrompt")
def is_token_prompt(prompt: PromptType) -> TypeIs[TokensPrompt]:
return isinstance(prompt, dict) and "prompt_token_ids" in prompt
def is_embeds_prompt(prompt: PromptType) -> TypeIs[EmbedsPrompt]:
return isinstance(prompt, dict) and "prompt_embeds" in prompt
def is_explicit_encoder_decoder_prompt(
prompt: PromptType) -> TypeIs[ExplicitEncoderDecoderPrompt]:
return isinstance(prompt, dict) and "encoder_prompt" in prompt
def is_embeds_inputs(inputs: SingletonInputs) -> TypeIs[EmbedsInputs]:
return isinstance(inputs, dict) and inputs["type"] == "embeds"
def split_enc_dec_inputs(
inputs: ProcessorInputs,
) -> tuple[Optional[SingletonInputs], SingletonInputs]: