chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[model-gateway] Add classification model support infrastructure (#16061)

Browse Source 
Co-authored-by: Chang Su <chang.s.su@oracle.com>
This commit is contained in:
Simo Lin
2025-12-29 08:34:05 -08:00
committed by GitHub
co-authored by Chang Su
parent 8e08207c18
commit 162d1cf9be
9 changed files with 167 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/grpc/sglang_scheduler.proto
+6
View File
@@ -428,6 +428,12 @@ message GetModelInfoResponse {
int32 bos_token_id = 13;
int32 max_req_input_len = 14;
repeated string architectures = 15;
// Classification model support (from HuggingFace config.json)
// id2label maps class indices to label names, e.g., {"0": "negative", "1": "positive"}
string id2label_json = 16;
// Number of classification labels (0 if not a classifier)
int32 num_labels = 17;
}
// Get server information
python/sglang/srt/grpc/sglang_scheduler_pb2.py
+8 -8
View File
File diff suppressed because one or more lines are too long
python/sglang/srt/grpc/sglang_scheduler_pb2.pyi
+6 -2
View File
@@ -432,7 +432,7 @@ class GetModelInfoRequest(_message.Message):
def __init__(self) -> None: ...
class GetModelInfoResponse(_message.Message):
__slots__ = ("model_path", "tokenizer_path", "is_generation", "preferred_sampling_params", "weight_version", "served_model_name", "max_context_length", "vocab_size", "supports_vision", "model_type", "eos_token_ids", "pad_token_id", "bos_token_id", "max_req_input_len", "architectures")
__slots__ = ("model_path", "tokenizer_path", "is_generation", "preferred_sampling_params", "weight_version", "served_model_name", "max_context_length", "vocab_size", "supports_vision", "model_type", "eos_token_ids", "pad_token_id", "bos_token_id", "max_req_input_len", "architectures", "id2label_json", "num_labels")
MODEL_PATH_FIELD_NUMBER: _ClassVar[int]
TOKENIZER_PATH_FIELD_NUMBER: _ClassVar[int]
IS_GENERATION_FIELD_NUMBER: _ClassVar[int]
@@ -448,6 +448,8 @@ class GetModelInfoResponse(_message.Message):
BOS_TOKEN_ID_FIELD_NUMBER: _ClassVar[int]
MAX_REQ_INPUT_LEN_FIELD_NUMBER: _ClassVar[int]
ARCHITECTURES_FIELD_NUMBER: _ClassVar[int]
ID2LABEL_JSON_FIELD_NUMBER: _ClassVar[int]
NUM_LABELS_FIELD_NUMBER: _ClassVar[int]
model_path: str
tokenizer_path: str
is_generation: bool
@@ -463,7 +465,9 @@ class GetModelInfoResponse(_message.Message):
bos_token_id: int
max_req_input_len: int
architectures: _containers.RepeatedScalarFieldContainer[str]
def __init__(self, model_path: _Optional[str] = ..., tokenizer_path: _Optional[str] = ..., is_generation: bool = ..., preferred_sampling_params: _Optional[str] = ..., weight_version: _Optional[str] = ..., served_model_name: _Optional[str] = ..., max_context_length: _Optional[int] = ..., vocab_size: _Optional[int] = ..., supports_vision: bool = ..., model_type: _Optional[str] = ..., eos_token_ids: _Optional[_Iterable[int]] = ..., pad_token_id: _Optional[int] = ..., bos_token_id: _Optional[int] = ..., max_req_input_len: _Optional[int] = ..., architectures: _Optional[_Iterable[str]] = ...) -> None: ...
id2label_json: str
num_labels: int
def __init__(self, model_path: _Optional[str] = ..., tokenizer_path: _Optional[str] = ..., is_generation: bool = ..., preferred_sampling_params: _Optional[str] = ..., weight_version: _Optional[str] = ..., served_model_name: _Optional[str] = ..., max_context_length: _Optional[int] = ..., vocab_size: _Optional[int] = ..., supports_vision: bool = ..., model_type: _Optional[str] = ..., eos_token_ids: _Optional[_Iterable[int]] = ..., pad_token_id: _Optional[int] = ..., bos_token_id: _Optional[int] = ..., max_req_input_len: _Optional[int] = ..., architectures: _Optional[_Iterable[str]] = ..., id2label_json: _Optional[str] = ..., num_labels: _Optional[int] = ...) -> None: ...
class GetServerInfoRequest(_message.Message):
__slots__ = ()