[Refactor][Frontend] Keep all logic about reasoning into one class (#14428)

Signed-off-by: Ce Gao <cegao@tensorchord.ai>
This commit is contained in:
Ce Gao
2025-03-28 00:23:30 -07:00
committed by GitHub
parent 2d9045fce8
commit 32b14baf8a
18 changed files with 171 additions and 200 deletions
+2 -1
View File
@@ -23,6 +23,7 @@ from vllm.executor.executor_base import ExecutorBase
from vllm.logger import init_logger
from vllm.model_executor.layers.quantization import QUANTIZATION_METHODS
from vllm.plugins import load_general_plugins
from vllm.reasoning import ReasoningParserManager
from vllm.test_utils import MODEL_WEIGHTS_S3_BUCKET, MODELS_ON_S3
from vllm.transformers_utils.utils import check_gguf_file
from vllm.usage.usage_lib import UsageContext
@@ -1119,7 +1120,7 @@ class EngineArgs:
parser.add_argument(
"--reasoning-parser",
type=str,
choices=["deepseek_r1", "granite"],
choices=list(ReasoningParserManager.reasoning_parsers),
default=None,
help=
"Select the reasoning parser depending on the model that you're "
+3 -2
View File
@@ -2080,8 +2080,9 @@ class LLMEngine:
guided_decoding.backend = guided_decoding.backend or \
self.decoding_config.guided_decoding_backend
logger.debug("Reasoning backend: %s",
self.decoding_config.reasoning_backend)
if self.decoding_config.reasoning_backend is not None:
logger.debug("Building with reasoning backend %s",
self.decoding_config.reasoning_backend)
processor = get_local_guided_decoding_logits_processor(
guided_params=guided_decoding,