Skip to content

vllm.renderers.grok2

logger module-attribute 

logger = init_logger(__name__)

Grok2Renderer

Bases: BaseRenderer

Source code in vllm/renderers/grok2.py 
 23
 24
 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
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
class Grok2Renderer(BaseRenderer):
    @classmethod
    def from_config(
        cls,
        config: ModelConfig,
        tokenizer_kwargs: dict[str, Any],
    ) -> "BaseRenderer":
        return cls(config, tokenizer_kwargs)

    def __init__(
        self,
        config: ModelConfig,
        tokenizer_kwargs: dict[str, Any],
    ) -> None:
        super().__init__(config)

        if config.skip_tokenizer_init:
            tokenizer = None
        else:
            tokenizer = cached_get_tokenizer(
                tokenizer_cls=Grok2Tokenizer,
                **tokenizer_kwargs,
            )

        self._tokenizer = tokenizer

    @property
    def tokenizer(self) -> Grok2Tokenizer | None:
        return self._tokenizer

    def get_tokenizer(self) -> Grok2Tokenizer:
        tokenizer = self.tokenizer
        if tokenizer is None:
            raise ValueError("Tokenizer not available when `skip_tokenizer_init=True`")

        return tokenizer

    def render_messages(
        self,
        messages: list[ChatCompletionMessageParam],
        params: ChatParams,
    ) -> tuple[list[ConversationMessage], TextPrompt | TokensPrompt | EmbedsPrompt]:
        tokenizer = self.get_tokenizer()
        conversation, mm_data, mm_uuids = parse_chat_messages(
            messages,
            self.config,
            content_format="string",
        )

        prompt_raw = tokenizer.apply_chat_template(
            conversation=conversation,
            messages=messages,
            **params.get_apply_chat_template_kwargs(),
        )

        prompt = self.render_completion(prompt_raw)
        if mm_data is not None:
            prompt["multi_modal_data"] = mm_data
        if mm_uuids is not None:
            prompt["multi_modal_uuids"] = mm_uuids

        return conversation, prompt

    async def render_messages_async(
        self,
        messages: list[ChatCompletionMessageParam],
        params: ChatParams,
    ) -> tuple[list[ConversationMessage], TextPrompt | TokensPrompt | EmbedsPrompt]:
        tokenizer = self.get_tokenizer()
        conversation, mm_data, mm_uuids = await parse_chat_messages_async(
            messages,
            self.config,
            content_format="string",
        )

        prompt_raw = tokenizer.apply_chat_template(
            conversation=conversation,
            messages=messages,
            **params.get_apply_chat_template_kwargs(),
        )

        prompt = self.render_completion(prompt_raw)
        if mm_data is not None:
            prompt["multi_modal_data"] = mm_data
        if mm_uuids is not None:
            prompt["multi_modal_uuids"] = mm_uuids

        return conversation, prompt

_tokenizer instance-attribute 

_tokenizer = tokenizer

tokenizer property 

tokenizer: Grok2Tokenizer | None

__init__ 

__init__(
    config: ModelConfig, tokenizer_kwargs: dict[str, Any]
) -> None
Source code in vllm/renderers/grok2.py 
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
def __init__(
    self,
    config: ModelConfig,
    tokenizer_kwargs: dict[str, Any],
) -> None:
    super().__init__(config)

    if config.skip_tokenizer_init:
        tokenizer = None
    else:
        tokenizer = cached_get_tokenizer(
            tokenizer_cls=Grok2Tokenizer,
            **tokenizer_kwargs,
        )

    self._tokenizer = tokenizer

from_config classmethod 

from_config(
    config: ModelConfig, tokenizer_kwargs: dict[str, Any]
) -> BaseRenderer
Source code in vllm/renderers/grok2.py 
24
25
26
27
28
29
30
@classmethod
def from_config(
    cls,
    config: ModelConfig,
    tokenizer_kwargs: dict[str, Any],
) -> "BaseRenderer":
    return cls(config, tokenizer_kwargs)

get_tokenizer 

get_tokenizer() -> Grok2Tokenizer
Source code in vllm/renderers/grok2.py 
53
54
55
56
57
58
def get_tokenizer(self) -> Grok2Tokenizer:
    tokenizer = self.tokenizer
    if tokenizer is None:
        raise ValueError("Tokenizer not available when `skip_tokenizer_init=True`")

    return tokenizer

render_messages 

render_messages(
    messages: list[ChatCompletionMessageParam],
    params: ChatParams,
) -> tuple[
    list[ConversationMessage],
    TextPrompt | TokensPrompt | EmbedsPrompt,
]
Source code in vllm/renderers/grok2.py 
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
def render_messages(
    self,
    messages: list[ChatCompletionMessageParam],
    params: ChatParams,
) -> tuple[list[ConversationMessage], TextPrompt | TokensPrompt | EmbedsPrompt]:
    tokenizer = self.get_tokenizer()
    conversation, mm_data, mm_uuids = parse_chat_messages(
        messages,
        self.config,
        content_format="string",
    )

    prompt_raw = tokenizer.apply_chat_template(
        conversation=conversation,
        messages=messages,
        **params.get_apply_chat_template_kwargs(),
    )

    prompt = self.render_completion(prompt_raw)
    if mm_data is not None:
        prompt["multi_modal_data"] = mm_data
    if mm_uuids is not None:
        prompt["multi_modal_uuids"] = mm_uuids

    return conversation, prompt

render_messages_async async 

render_messages_async(
    messages: list[ChatCompletionMessageParam],
    params: ChatParams,
) -> tuple[
    list[ConversationMessage],
    TextPrompt | TokensPrompt | EmbedsPrompt,
]
Source code in vllm/renderers/grok2.py 
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
async def render_messages_async(
    self,
    messages: list[ChatCompletionMessageParam],
    params: ChatParams,
) -> tuple[list[ConversationMessage], TextPrompt | TokensPrompt | EmbedsPrompt]:
    tokenizer = self.get_tokenizer()
    conversation, mm_data, mm_uuids = await parse_chat_messages_async(
        messages,
        self.config,
        content_format="string",
    )

    prompt_raw = tokenizer.apply_chat_template(
        conversation=conversation,
        messages=messages,
        **params.get_apply_chat_template_kwargs(),
    )

    prompt = self.render_completion(prompt_raw)
    if mm_data is not None:
        prompt["multi_modal_data"] = mm_data
    if mm_uuids is not None:
        prompt["multi_modal_uuids"] = mm_uuids

    return conversation, prompt