[Refactor] Clean up pooling models build_tok_params logic (#42341)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye
2026-05-12 15:05:05 +00:00
committed by GitHub
parent 418ba8ef14
commit a1b2d87498
5 changed files with 115 additions and 136 deletions
+10 -35
View File
@@ -18,6 +18,7 @@ from ..base.protocol import (
CompletionRequestMixin,
EmbedRequestMixin,
EncodingRequestMixin,
FixedMaxLenTokenizeParamsMixin,
PoolingBasicRequestMixin,
)
@@ -27,22 +28,10 @@ class PoolingCompletionRequest(
CompletionRequestMixin,
EmbedRequestMixin,
ClassifyRequestMixin,
FixedMaxLenTokenizeParamsMixin,
):
task: PoolingTask | None = None
def build_tok_params(self, model_config: ModelConfig) -> TokenizeParams:
encoder_config = model_config.encoder_config or {}
return TokenizeParams(
max_total_tokens=model_config.max_model_len,
max_output_tokens=0,
truncate_prompt_tokens=self.truncate_prompt_tokens,
truncation_side=self.truncation_side,
do_lower_case=encoder_config.get("do_lower_case", False),
add_special_tokens=self.add_special_tokens,
max_total_tokens_param="max_model_len",
)
def to_pooling_params(self):
return PoolingParams(
task=self.task,
@@ -52,23 +41,14 @@ class PoolingCompletionRequest(
class PoolingChatRequest(
PoolingBasicRequestMixin, ChatRequestMixin, EmbedRequestMixin, ClassifyRequestMixin
PoolingBasicRequestMixin,
ChatRequestMixin,
EmbedRequestMixin,
ClassifyRequestMixin,
FixedMaxLenTokenizeParamsMixin,
):
task: PoolingTask | None = None
def build_tok_params(self, model_config: ModelConfig) -> TokenizeParams:
encoder_config = model_config.encoder_config or {}
return TokenizeParams(
max_total_tokens=model_config.max_model_len,
max_output_tokens=0,
truncate_prompt_tokens=self.truncate_prompt_tokens,
truncation_side=self.truncation_side,
do_lower_case=encoder_config.get("do_lower_case", False),
add_special_tokens=self.add_special_tokens,
max_total_tokens_param="max_model_len",
)
def to_pooling_params(self):
return PoolingParams(
task=self.task,
@@ -85,16 +65,11 @@ class IOProcessorRequest(PoolingBasicRequestMixin, EncodingRequestMixin, Generic
task: PoolingTask = "plugin"
def build_tok_params(self, model_config: ModelConfig) -> TokenizeParams:
encoder_config = model_config.encoder_config or {}
return TokenizeParams(
return self._build_pooling_tok_params(
model_config,
add_special_tokens=not model_config.is_encoder_decoder,
max_total_tokens=model_config.max_model_len,
max_output_tokens=0,
truncate_prompt_tokens=self.truncate_prompt_tokens,
truncation_side=self.truncation_side,
do_lower_case=encoder_config.get("do_lower_case", False),
add_special_tokens=not model_config.is_encoder_decoder,
max_total_tokens_param="max_model_len",
)
def to_pooling_params(self):