chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 14 Packages Projects Releases Wiki Activity

Remove dp balance metadata and minimul token balance. (#11170)

Browse Source
This commit is contained in:
Liangsheng Yin
2025-10-03 01:48:15 +08:00
committed by GitHub
parent bfcd9b2433
commit 7ff740a6ce
11 changed files with 77 additions and 316 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/grpc/sglang_scheduler.proto
+1 -4
View File
@@ -120,11 +120,8 @@ message GenerateRequest {
// Data parallel routing
int32 data_parallel_rank = 16;
// For load balancing
int32 dp_balance_id = 17;
// Whether client wants streaming response
bool stream = 18;
bool stream = 17;
}
message TokenizedInput {
python/sglang/srt/grpc/sglang_scheduler_pb2.py
+64 -64
View File
File diff suppressed because one or more lines are too long
python/sglang/srt/grpc/sglang_scheduler_pb2.pyi
+2 -4
View File
@@ -82,7 +82,7 @@ class DisaggregatedParams(_message.Message):
def __init__(self, bootstrap_host: _Optional[str] = ..., bootstrap_port: _Optional[int] = ..., bootstrap_room: _Optional[int] = ...) -> None: ...
class GenerateRequest(_message.Message):
__slots__ = ("request_id", "tokenized", "mm_inputs", "sampling_params", "return_logprob", "logprob_start_len", "top_logprobs_num", "token_ids_logprob", "return_hidden_states", "disaggregated_params", "custom_logit_processor", "timestamp", "log_metrics", "input_embeds", "lora_id", "data_parallel_rank", "dp_balance_id", "stream")
__slots__ = ("request_id", "tokenized", "mm_inputs", "sampling_params", "return_logprob", "logprob_start_len", "top_logprobs_num", "token_ids_logprob", "return_hidden_states", "disaggregated_params", "custom_logit_processor", "timestamp", "log_metrics", "input_embeds", "lora_id", "data_parallel_rank", "stream")
REQUEST_ID_FIELD_NUMBER: _ClassVar[int]
TOKENIZED_FIELD_NUMBER: _ClassVar[int]
MM_INPUTS_FIELD_NUMBER: _ClassVar[int]
@@ -99,7 +99,6 @@ class GenerateRequest(_message.Message):
INPUT_EMBEDS_FIELD_NUMBER: _ClassVar[int]
LORA_ID_FIELD_NUMBER: _ClassVar[int]
DATA_PARALLEL_RANK_FIELD_NUMBER: _ClassVar[int]
DP_BALANCE_ID_FIELD_NUMBER: _ClassVar[int]
STREAM_FIELD_NUMBER: _ClassVar[int]
request_id: str
tokenized: TokenizedInput
@@ -117,9 +116,8 @@ class GenerateRequest(_message.Message):
input_embeds: _containers.RepeatedScalarFieldContainer[float]
lora_id: str
data_parallel_rank: int
dp_balance_id: int
stream: bool
def __init__(self, request_id: _Optional[str] = ..., tokenized: _Optional[_Union[TokenizedInput, _Mapping]] = ..., mm_inputs: _Optional[_Union[MultimodalInputs, _Mapping]] = ..., sampling_params: _Optional[_Union[SamplingParams, _Mapping]] = ..., return_logprob: bool = ..., logprob_start_len: _Optional[int] = ..., top_logprobs_num: _Optional[int] = ..., token_ids_logprob: _Optional[_Iterable[int]] = ..., return_hidden_states: bool = ..., disaggregated_params: _Optional[_Union[DisaggregatedParams, _Mapping]] = ..., custom_logit_processor: _Optional[str] = ..., timestamp: _Optional[_Union[datetime.datetime, _timestamp_pb2.Timestamp, _Mapping]] = ..., log_metrics: bool = ..., input_embeds: _Optional[_Iterable[float]] = ..., lora_id: _Optional[str] = ..., data_parallel_rank: _Optional[int] = ..., dp_balance_id: _Optional[int] = ..., stream: bool = ...) -> None: ...
def __init__(self, request_id: _Optional[str] = ..., tokenized: _Optional[_Union[TokenizedInput, _Mapping]] = ..., mm_inputs: _Optional[_Union[MultimodalInputs, _Mapping]] = ..., sampling_params: _Optional[_Union[SamplingParams, _Mapping]] = ..., return_logprob: bool = ..., logprob_start_len: _Optional[int] = ..., top_logprobs_num: _Optional[int] = ..., token_ids_logprob: _Optional[_Iterable[int]] = ..., return_hidden_states: bool = ..., disaggregated_params: _Optional[_Union[DisaggregatedParams, _Mapping]] = ..., custom_logit_processor: _Optional[str] = ..., timestamp: _Optional[_Union[datetime.datetime, _timestamp_pb2.Timestamp, _Mapping]] = ..., log_metrics: bool = ..., input_embeds: _Optional[_Iterable[float]] = ..., lora_id: _Optional[str] = ..., data_parallel_rank: _Optional[int] = ..., stream: bool = ...) -> None: ...
class TokenizedInput(_message.Message):
__slots__ = ("original_text", "input_ids")