Consolidate rendering parameters into RenderConfig dataclass (#24543)

Signed-off-by: sfeng33 <4florafeng@gmail.com>
This commit is contained in:
Flora Feng
2025-09-10 08:44:47 +00:00
committed by GitHub
parent feaf202e93
commit 77f62613f9
8 changed files with 167 additions and 108 deletions
+15 -1
View File
@@ -58,7 +58,8 @@ from vllm.entrypoints.openai.protocol import (ChatCompletionRequest,
TranslationRequest)
from vllm.entrypoints.openai.serving_models import OpenAIServingModels
from vllm.entrypoints.openai.tool_parsers import ToolParser
from vllm.entrypoints.renderer import BaseRenderer, CompletionRenderer
from vllm.entrypoints.renderer import (BaseRenderer, CompletionRenderer,
RenderConfig)
# yapf: enable
from vllm.inputs.data import PromptType
from vllm.inputs.data import TokensPrompt as EngineTokensPrompt
@@ -248,6 +249,19 @@ class OpenAIServing:
tokenizer=tokenizer,
async_tokenizer_pool=self._async_tokenizer_pool)
def _build_render_config(
self,
request: Any,
) -> RenderConfig:
"""
Build and return a `RenderConfig` for an endpoint.
Used by the renderer to control how prompts are prepared
(e.g., tokenization and length handling). Endpoints should
implement this with logic appropriate to their request type.
"""
raise NotImplementedError
def _get_async_tokenizer(self, tokenizer) -> AsyncMicrobatchTokenizer:
"""
Return (and cache) an `AsyncMicrobatchTokenizer` bound to the