wehub-resource-sync
66 lines
2.0 KiB
Python
66 lines
2.0 KiB
Python
# SPDX-License-Identifier: Apache-2.0
|
|
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
|
|
|
from vllm.entrypoints.chat_utils import (
|
|
ChatCompletionMessageParam,
|
|
ConversationMessage,
|
|
parse_chat_messages,
|
|
parse_chat_messages_async,
|
|
)
|
|
from vllm.logger import init_logger
|
|
|
|
from .base import BaseRenderer
|
|
from .inputs import DictPrompt
|
|
from .inputs.preprocess import parse_dec_only_prompt
|
|
from .params import ChatParams
|
|
|
|
logger = init_logger(__name__)
|
|
|
|
|
|
class TerratorchRenderer(BaseRenderer):
|
|
def render_messages(
|
|
self,
|
|
messages: list[ChatCompletionMessageParam],
|
|
params: ChatParams,
|
|
) -> tuple[list[ConversationMessage], DictPrompt]:
|
|
model_config = self.model_config
|
|
|
|
conversation, mm_data, mm_uuids = parse_chat_messages(
|
|
messages,
|
|
model_config,
|
|
content_format="string",
|
|
media_io_kwargs=params.media_io_kwargs,
|
|
mm_processor_kwargs=params.mm_processor_kwargs,
|
|
)
|
|
|
|
prompt = parse_dec_only_prompt([1]) # Dummy token IDs
|
|
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], DictPrompt]:
|
|
model_config = self.model_config
|
|
|
|
conversation, mm_data, mm_uuids = await parse_chat_messages_async(
|
|
messages,
|
|
model_config,
|
|
content_format="string",
|
|
media_io_kwargs=params.media_io_kwargs,
|
|
mm_processor_kwargs=params.mm_processor_kwargs,
|
|
)
|
|
|
|
prompt = parse_dec_only_prompt([1]) # Dummy token IDs
|
|
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
|