[Frontend] Use new Renderer for Completions and Tokenize API (#32863)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
@@ -3,10 +3,9 @@
|
||||
import time
|
||||
from typing import Any, TypeAlias
|
||||
|
||||
from pydantic import (
|
||||
Field,
|
||||
)
|
||||
from pydantic import Field
|
||||
|
||||
from vllm.config import ModelConfig
|
||||
from vllm.entrypoints.openai.engine.protocol import OpenAIBaseModel, UsageInfo
|
||||
from vllm.entrypoints.pooling.base.protocol import (
|
||||
ChatRequestMixin,
|
||||
@@ -14,15 +13,47 @@ from vllm.entrypoints.pooling.base.protocol import (
|
||||
EmbedRequestMixin,
|
||||
PoolingBasicRequestMixin,
|
||||
)
|
||||
from vllm.renderers import TokenizeParams
|
||||
from vllm.utils import random_uuid
|
||||
|
||||
|
||||
def _get_max_total_output_tokens(
|
||||
model_config: ModelConfig,
|
||||
) -> tuple[int | None, int]:
|
||||
max_total_tokens = model_config.max_model_len
|
||||
pooler_config = model_config.pooler_config
|
||||
|
||||
if pooler_config is None:
|
||||
return max_total_tokens, 0
|
||||
|
||||
if pooler_config.enable_chunked_processing:
|
||||
return None, 0
|
||||
|
||||
max_embed_len = pooler_config.max_embed_len or max_total_tokens
|
||||
max_output_tokens = max_total_tokens - max_embed_len
|
||||
return max_total_tokens, max_output_tokens
|
||||
|
||||
|
||||
class EmbeddingCompletionRequest(
|
||||
PoolingBasicRequestMixin, CompletionRequestMixin, EmbedRequestMixin
|
||||
):
|
||||
# Ordered by official OpenAI API documentation
|
||||
# https://platform.openai.com/docs/api-reference/embeddings
|
||||
pass
|
||||
def build_tok_params(self, model_config: ModelConfig) -> TokenizeParams:
|
||||
encoder_config = model_config.encoder_config or {}
|
||||
|
||||
(
|
||||
max_total_tokens,
|
||||
max_output_tokens,
|
||||
) = _get_max_total_output_tokens(model_config)
|
||||
|
||||
return TokenizeParams(
|
||||
max_total_tokens=max_total_tokens,
|
||||
max_output_tokens=max_output_tokens,
|
||||
truncate_prompt_tokens=self.truncate_prompt_tokens,
|
||||
do_lower_case=encoder_config.get("do_lower_case", False),
|
||||
add_special_tokens=self.add_special_tokens,
|
||||
max_total_tokens_param="max_model_len",
|
||||
max_output_tokens_param="max_model_len - max_embed_len",
|
||||
)
|
||||
|
||||
|
||||
class EmbeddingChatRequest(
|
||||
@@ -33,6 +64,24 @@ class EmbeddingChatRequest(
|
||||
description=("Additional kwargs to pass to the HF processor."),
|
||||
)
|
||||
|
||||
def build_tok_params(self, model_config: ModelConfig) -> TokenizeParams:
|
||||
encoder_config = model_config.encoder_config or {}
|
||||
|
||||
(
|
||||
max_total_tokens,
|
||||
max_output_tokens,
|
||||
) = _get_max_total_output_tokens(model_config)
|
||||
|
||||
return TokenizeParams(
|
||||
max_total_tokens=max_total_tokens,
|
||||
max_output_tokens=max_output_tokens,
|
||||
truncate_prompt_tokens=self.truncate_prompt_tokens,
|
||||
do_lower_case=encoder_config.get("do_lower_case", False),
|
||||
add_special_tokens=self.add_special_tokens,
|
||||
max_total_tokens_param="max_model_len",
|
||||
max_output_tokens_param="max_model_len - max_embed_len",
|
||||
)
|
||||
|
||||
|
||||
EmbeddingRequest: TypeAlias = EmbeddingCompletionRequest | EmbeddingChatRequest
|
||||
|
||||
|
||||
Reference in New Issue
Block a user