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
+17 -8
View File
@@ -30,6 +30,7 @@ from vllm.entrypoints.openai.serving_engine import (OpenAIServing,
clamp_prompt_logprobs)
# yapf: enable
from vllm.entrypoints.openai.serving_models import OpenAIServingModels
from vllm.entrypoints.renderer import RenderConfig
from vllm.entrypoints.utils import get_max_tokens
from vllm.inputs.data import (EmbedsPrompt, TokensPrompt, is_embeds_prompt,
is_tokens_prompt)
@@ -129,18 +130,11 @@ class OpenAIServingCompletion(OpenAIServing):
tokenizer = await self.engine_client.get_tokenizer(lora_request
)
renderer = self._get_renderer(tokenizer)
max_input_tokens_len = self.max_model_len - (request.max_tokens
or 0)
engine_prompts = await renderer.render_prompt_and_embeds(
prompt_or_prompts=request.prompt,
prompt_embeds=request.prompt_embeds,
max_length=max_input_tokens_len,
truncate_prompt_tokens=request.truncate_prompt_tokens,
add_special_tokens=request.add_special_tokens,
cache_salt=request.cache_salt,
needs_detokenization=bool(request.echo
and not request.return_token_ids),
config=self._build_render_config(request),
)
except ValueError as e:
logger.exception("Error in preprocessing prompt inputs")
@@ -677,3 +671,18 @@ class OpenAIServingCompletion(OpenAIServing):
tokens=out_tokens,
top_logprobs=out_top_logprobs,
)
def _build_render_config(
self,
request: CompletionRequest,
max_input_length: Optional[int] = None,
) -> RenderConfig:
max_input_tokens_len = self.max_model_len - (request.max_tokens or 0)
return RenderConfig(
max_length=max_input_tokens_len,
truncate_prompt_tokens=request.truncate_prompt_tokens,
add_special_tokens=request.add_special_tokens,
cache_salt=request.cache_salt,
needs_detokenization=bool(request.echo
and not request.return_token_ids),
)