Skip to content

vllm.renderers.inputs.preprocess

Schemas and utilities for preprocessing inputs.

Classes:

Functions:

Attributes:

DecoderDictPrompt = TextPrompt | TokensPrompt module-attribute

A DecoderPrompt that has been standardized into a dictionary.

DecoderOnlyDictPrompt = TextPrompt | TokensPrompt | EmbedsPrompt module-attribute

A DecoderOnlyPrompt that has been standardized into a dictionary.

DictPrompt = DecoderOnlyDictPrompt | EncoderDecoderDictPrompt module-attribute

A PromptType that has been standardized into a dictionary.

EncoderDictPrompt = TextPrompt | TokensPrompt module-attribute

A EncoderPrompt that has been standardized into a dictionary.

SingletonDictPrompt = DecoderOnlyDictPrompt | EncoderDictPrompt | DecoderDictPrompt module-attribute

A SingletonPrompt that has been standardized into a dictionary.

EncoderDecoderDictPrompt

Bases: TypedDict

A EncoderDecoderPrompt that has been standardized into a dictionary.

Source code in vllm/renderers/inputs/preprocess.py 
 92
 93
 94
 95
 96
 97
 98
 99
100
class EncoderDecoderDictPrompt(TypedDict):
    """
    A [`EncoderDecoderPrompt`][vllm.inputs.llm.EncoderDecoderPrompt]
    that has been standardized into a dictionary.
    """

    encoder_prompt: EncoderDictPrompt

    decoder_prompt: DecoderDictPrompt | None

_validate_prompt_dict(prompt)

Reject malformed dict prompts before renderer tokenization.

Source code in vllm/renderers/inputs/preprocess.py 
119
120
121
122
123
124
125
126
127
128
129
def _validate_prompt_dict(prompt: Mapping[str, object]) -> None:
    """Reject malformed dict prompts before renderer tokenization."""
    if (
        "prompt" not in prompt
        or "prompt_token_ids" in prompt
        or "prompt_embeds" in prompt
    ):
        return

    if not isinstance(prompt["prompt"], str):
        raise TypeError("Prompt text should be a string")

parse_dec_only_prompt(prompt)

Parse a prompt for a decoder-only model and normalize it to a dictionary.

Source code in vllm/renderers/inputs/preprocess.py 
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
def parse_dec_only_prompt(prompt: PromptType | object) -> DecoderOnlyDictPrompt:
    """
    Parse a prompt for a decoder-only model and normalize it to a dictionary.
    """
    if isinstance(prompt, str):
        return TextPrompt(prompt=prompt)

    if isinstance(prompt, list):
        if not is_list_of(prompt, int):
            raise TypeError("Token prompt should be a list of integers")

        return TokensPrompt(prompt_token_ids=prompt)

    if isinstance(prompt, dict):
        if "encoder_prompt" in prompt:
            raise TypeError("Cannot pass encoder-decoder prompt to decoder-only models")

        _validate_prompt_dict(prompt)

        if (
            "prompt" in prompt
            or "prompt_token_ids" in prompt
            or "prompt_embeds" in prompt
        ):
            return prompt  # type: ignore[return-value]

        raise TypeError("Prompt dictionary must contain text, tokens, or embeddings")

    raise TypeError("Prompt should be a string, list of tokens, or dictionary")

parse_enc_dec_prompt(prompt)

Parse a prompt for an encoder-decoder model and normalize it to a dictionary.

Source code in vllm/renderers/inputs/preprocess.py 
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
def parse_enc_dec_prompt(prompt: PromptType | object) -> EncoderDecoderDictPrompt:
    """
    Parse a prompt for an encoder-decoder model and normalize it to a dictionary.
    """
    if isinstance(prompt, dict) and "encoder_prompt" in prompt:
        enc_prompt = prompt["encoder_prompt"]  # type: ignore[typeddict-item]
        dec_prompt = prompt["decoder_prompt"]  # type: ignore[typeddict-item]
    else:
        enc_prompt = prompt
        dec_prompt = None

    return EncoderDecoderDictPrompt(
        encoder_prompt=_parse_enc_prompt(enc_prompt),
        decoder_prompt=None if dec_prompt is None else _parse_dec_prompt(dec_prompt),
    )