Skip to content

vllm.inputs.parse

ParsedSingletonPrompt module-attribute 

ParsedSingletonPrompt: TypeAlias = (
    ParsedStrPrompt
    | ParsedTextPrompt
    | ParsedTokensPrompt
    | ParsedEmbedsPrompt
)

ParsedEmbedsPrompt

Bases: TypedDict

Source code in vllm/inputs/parse.py 
74
75
76
class ParsedEmbedsPrompt(TypedDict):
    type: Literal["embeds"]
    content: EmbedsPrompt

content instance-attribute 

content: EmbedsPrompt

type instance-attribute 

type: Literal['embeds']

ParsedStrPrompt

Bases: TypedDict

Source code in vllm/inputs/parse.py 
59
60
61
class ParsedStrPrompt(TypedDict):
    type: Literal["str"]
    content: str

content instance-attribute 

content: str

type instance-attribute 

type: Literal['str']

ParsedTextPrompt

Bases: TypedDict

Source code in vllm/inputs/parse.py 
64
65
66
class ParsedTextPrompt(TypedDict):
    type: Literal["text"]
    content: TextPrompt

content instance-attribute 

content: TextPrompt

type instance-attribute 

type: Literal['text']

ParsedTokensPrompt

Bases: TypedDict

Source code in vllm/inputs/parse.py 
69
70
71
class ParsedTokensPrompt(TypedDict):
    type: Literal["tokens"]
    content: TokensPrompt

content instance-attribute 

content: TokensPrompt

type instance-attribute 

type: Literal['tokens']

PromptComponents

Bases: NamedTuple

Source code in vllm/inputs/parse.py 
120
121
122
123
class PromptComponents(NamedTuple):
    text: str | None = None
    token_ids: list[int] | None = None
    embeds: "torch.Tensor | None" = None

embeds class-attribute instance-attribute 

embeds: Tensor | None = None

text class-attribute instance-attribute 

text: str | None = None

token_ids class-attribute instance-attribute 

token_ids: list[int] | None = None

get_prompt_components 

get_prompt_components(
    prompt: PromptType,
) -> PromptComponents
Source code in vllm/inputs/parse.py 
126
127
128
129
130
131
132
133
134
135
136
137
def get_prompt_components(prompt: PromptType) -> PromptComponents:
    if isinstance(prompt, str):
        return PromptComponents(text=prompt)

    if encoder_prompt := prompt.get("encoder_prompt"):
        return get_prompt_components(encoder_prompt)  # type: ignore[arg-type]

    return PromptComponents(
        text=prompt.get("prompt"),  # type: ignore[arg-type]
        token_ids=prompt.get("prompt_token_ids"),  # type: ignore[arg-type]
        embeds=prompt.get("prompt_embeds"),
    )

is_explicit_encoder_decoder_prompt 

is_explicit_encoder_decoder_prompt(
    prompt: PromptType,
) -> TypeIs[ExplicitEncoderDecoderPrompt]
Source code in vllm/inputs/parse.py 
101
102
103
104
def is_explicit_encoder_decoder_prompt(
    prompt: PromptType,
) -> TypeIs[ExplicitEncoderDecoderPrompt]:
    return isinstance(prompt, dict) and "encoder_prompt" in prompt

parse_raw_prompts 

parse_raw_prompts(
    prompt: str | list[str] | list[int] | list[list[int]],
) -> Sequence[TextPrompt] | Sequence[TokensPrompt]
Source code in vllm/inputs/parse.py 
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
def parse_raw_prompts(
    prompt: str | list[str] | list[int] | list[list[int]],
) -> Sequence[TextPrompt] | Sequence[TokensPrompt]:
    if isinstance(prompt, str):
        # case 1: a string
        return [TextPrompt(prompt=prompt)]

    if isinstance(prompt, list):
        if len(prompt) == 0:
            raise ValueError("please provide at least one prompt")

        if is_list_of(prompt, str):
            # case 2: array of strings
            prompt = cast(list[str], prompt)
            return [TextPrompt(prompt=elem) for elem in prompt]
        if is_list_of(prompt, int):
            # case 3: array of tokens
            prompt = cast(list[int], prompt)
            return [TokensPrompt(prompt_token_ids=prompt)]
        if is_list_of(prompt, list):
            prompt = cast(list[list[int]], prompt)
            if len(prompt[0]) == 0:
                raise ValueError("please provide at least one prompt")

            if is_list_of(prompt[0], int):
                # case 4: array of token arrays
                return [TokensPrompt(prompt_token_ids=elem) for elem in prompt]

    raise TypeError(
        "prompt must be a string, array of strings, "
        "array of tokens, or array of token arrays"
    )

parse_singleton_prompt 

parse_singleton_prompt(
    prompt: SingletonPrompt,
) -> ParsedSingletonPrompt
Source code in vllm/inputs/parse.py 
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
def parse_singleton_prompt(prompt: SingletonPrompt) -> ParsedSingletonPrompt:
    if isinstance(prompt, str):
        return ParsedStrPrompt(type="str", content=prompt)
    elif isinstance(prompt, dict):
        # 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, TokensPrompt, or EmbedsPrompt"
    )

split_enc_dec_inputs 

split_enc_dec_inputs(
    inputs: ProcessorInputs,
) -> tuple[SingletonInputs | None, SingletonInputs]
Source code in vllm/inputs/parse.py 
107
108
109
110
111
112
113
114
115
116
117
def split_enc_dec_inputs(
    inputs: ProcessorInputs,
) -> tuple[SingletonInputs | None, SingletonInputs]:
    if "encoder" in inputs and "decoder" in inputs:
        # NOTE: This passes pyright but not mypy
        return (
            inputs["encoder"],  # type: ignore[typeddict-item]
            inputs["decoder"],  # type: ignore[typeddict-item]
        )

    return None, inputs