chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity
Files
cc63c99f112f781605017db018fced686cdc94de
sglang/sgl-model-gateway/bindings/golang/internal/proto/sglang_scheduler.pb.go
T

3326 lines
104 KiB
Go
Raw Blame History

// Code generated by protoc-gen-go. DO NOT EDIT.
// versions:
// protoc-gen-go v1.36.10
// protoc v3.21.12
// source: sglang_scheduler.proto
package proto
import (
protoreflect "google.golang.org/protobuf/reflect/protoreflect"
protoimpl "google.golang.org/protobuf/runtime/protoimpl"
structpb "google.golang.org/protobuf/types/known/structpb"
timestamppb "google.golang.org/protobuf/types/known/timestamppb"
reflect "reflect"
sync "sync"
unsafe "unsafe"
)
const (
// Verify that this generated code is sufficiently up-to-date.
_ = protoimpl.EnforceVersion(20 - protoimpl.MinVersion)
// Verify that runtime/protoimpl is sufficiently up-to-date.
_ = protoimpl.EnforceVersion(protoimpl.MaxVersion - 20)
)
// Sampling parameters matching SGLang's SamplingParams
//
// IMPORTANT: Do not use SamplingParams::default() directly!
// The proto3 defaults (0 for numeric fields) do NOT match the semantic defaults
// (temperature=1.0, top_p=1.0, top_k=-1, etc.). Always construct with explicit values
// or use the conversion functions in sglang_scheduler.rs / grpc_server.py.
type SamplingParams struct {
state protoimpl.MessageState `protogen:"open.v1"`
Temperature float32 `protobuf:"fixed32,1,opt,name=temperature,proto3" json:"temperature,omitempty"`
TopP float32 `protobuf:"fixed32,2,opt,name=top_p,json=topP,proto3" json:"top_p,omitempty"`
TopK int32 `protobuf:"varint,3,opt,name=top_k,json=topK,proto3" json:"top_k,omitempty"`
MinP float32 `protobuf:"fixed32,4,opt,name=min_p,json=minP,proto3" json:"min_p,omitempty"`
FrequencyPenalty float32 `protobuf:"fixed32,5,opt,name=frequency_penalty,json=frequencyPenalty,proto3" json:"frequency_penalty,omitempty"`
PresencePenalty float32 `protobuf:"fixed32,6,opt,name=presence_penalty,json=presencePenalty,proto3" json:"presence_penalty,omitempty"`
RepetitionPenalty float32 `protobuf:"fixed32,7,opt,name=repetition_penalty,json=repetitionPenalty,proto3" json:"repetition_penalty,omitempty"`
MaxNewTokens *int32 `protobuf:"varint,8,opt,name=max_new_tokens,json=maxNewTokens,proto3,oneof" json:"max_new_tokens,omitempty"`
Stop []string `protobuf:"bytes,9,rep,name=stop,proto3" json:"stop,omitempty"`
StopTokenIds []uint32 `protobuf:"varint,10,rep,packed,name=stop_token_ids,json=stopTokenIds,proto3" json:"stop_token_ids,omitempty"`
SkipSpecialTokens bool `protobuf:"varint,11,opt,name=skip_special_tokens,json=skipSpecialTokens,proto3" json:"skip_special_tokens,omitempty"`
SpacesBetweenSpecialTokens bool `protobuf:"varint,12,opt,name=spaces_between_special_tokens,json=spacesBetweenSpecialTokens,proto3" json:"spaces_between_special_tokens,omitempty"`
// Structured generation
//
// Types that are valid to be assigned to Constraint:
//
// *SamplingParams_Regex
// *SamplingParams_JsonSchema
// *SamplingParams_EbnfGrammar
// *SamplingParams_StructuralTag
Constraint isSamplingParams_Constraint `protobuf_oneof:"constraint"`
// Speculative decoding
N int32 `protobuf:"varint,17,opt,name=n,proto3" json:"n,omitempty"` // Number of samples
// Additional parameters
MinNewTokens int32 `protobuf:"varint,18,opt,name=min_new_tokens,json=minNewTokens,proto3" json:"min_new_tokens,omitempty"`
IgnoreEos bool `protobuf:"varint,19,opt,name=ignore_eos,json=ignoreEos,proto3" json:"ignore_eos,omitempty"`
NoStopTrim bool `protobuf:"varint,20,opt,name=no_stop_trim,json=noStopTrim,proto3" json:"no_stop_trim,omitempty"`
StreamInterval *int32 `protobuf:"varint,21,opt,name=stream_interval,json=streamInterval,proto3,oneof" json:"stream_interval,omitempty"`
LogitBias map[string]float32 `protobuf:"bytes,22,rep,name=logit_bias,json=logitBias,proto3" json:"logit_bias,omitempty" protobuf_key:"bytes,1,opt,name=key" protobuf_val:"fixed32,2,opt,name=value"`
// Custom parameters for extensibility
CustomParams *structpb.Struct `protobuf:"bytes,23,opt,name=custom_params,json=customParams,proto3" json:"custom_params,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *SamplingParams) Reset() {
*x = SamplingParams{}
mi := &file_sglang_scheduler_proto_msgTypes[0]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *SamplingParams) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*SamplingParams) ProtoMessage() {}
func (x *SamplingParams) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[0]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use SamplingParams.ProtoReflect.Descriptor instead.
func (*SamplingParams) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{0}
}
func (x *SamplingParams) GetTemperature() float32 {
if x != nil {
return x.Temperature
}
return 0
}
func (x *SamplingParams) GetTopP() float32 {
if x != nil {
return x.TopP
}
return 0
}
func (x *SamplingParams) GetTopK() int32 {
if x != nil {
return x.TopK
}
return 0
}
func (x *SamplingParams) GetMinP() float32 {
if x != nil {
return x.MinP
}
return 0
}
func (x *SamplingParams) GetFrequencyPenalty() float32 {
if x != nil {
return x.FrequencyPenalty
}
return 0
}
func (x *SamplingParams) GetPresencePenalty() float32 {
if x != nil {
return x.PresencePenalty
}
return 0
}
func (x *SamplingParams) GetRepetitionPenalty() float32 {
if x != nil {
return x.RepetitionPenalty
}
return 0
}
func (x *SamplingParams) GetMaxNewTokens() int32 {
if x != nil && x.MaxNewTokens != nil {
return *x.MaxNewTokens
}
return 0
}
func (x *SamplingParams) GetStop() []string {
if x != nil {
return x.Stop
}
return nil
}
func (x *SamplingParams) GetStopTokenIds() []uint32 {
if x != nil {
return x.StopTokenIds
}
return nil
}
func (x *SamplingParams) GetSkipSpecialTokens() bool {
if x != nil {
return x.SkipSpecialTokens
}
return false
}
func (x *SamplingParams) GetSpacesBetweenSpecialTokens() bool {
if x != nil {
return x.SpacesBetweenSpecialTokens
}
return false
}
func (x *SamplingParams) GetConstraint() isSamplingParams_Constraint {
if x != nil {
return x.Constraint
}
return nil
}
func (x *SamplingParams) GetRegex() string {
if x != nil {
if x, ok := x.Constraint.(*SamplingParams_Regex); ok {
return x.Regex
}
}
return ""
}
func (x *SamplingParams) GetJsonSchema() string {
if x != nil {
if x, ok := x.Constraint.(*SamplingParams_JsonSchema); ok {
return x.JsonSchema
}
}
return ""
}
func (x *SamplingParams) GetEbnfGrammar() string {
if x != nil {
if x, ok := x.Constraint.(*SamplingParams_EbnfGrammar); ok {
return x.EbnfGrammar
}
}
return ""
}
func (x *SamplingParams) GetStructuralTag() string {
if x != nil {
if x, ok := x.Constraint.(*SamplingParams_StructuralTag); ok {
return x.StructuralTag
}
}
return ""
}
func (x *SamplingParams) GetN() int32 {
if x != nil {
return x.N
}
return 0
}
func (x *SamplingParams) GetMinNewTokens() int32 {
if x != nil {
return x.MinNewTokens
}
return 0
}
func (x *SamplingParams) GetIgnoreEos() bool {
if x != nil {
return x.IgnoreEos
}
return false
}
func (x *SamplingParams) GetNoStopTrim() bool {
if x != nil {
return x.NoStopTrim
}
return false
}
func (x *SamplingParams) GetStreamInterval() int32 {
if x != nil && x.StreamInterval != nil {
return *x.StreamInterval
}
return 0
}
func (x *SamplingParams) GetLogitBias() map[string]float32 {
if x != nil {
return x.LogitBias
}
return nil
}
func (x *SamplingParams) GetCustomParams() *structpb.Struct {
if x != nil {
return x.CustomParams
}
return nil
}
type isSamplingParams_Constraint interface {
isSamplingParams_Constraint()
}
type SamplingParams_Regex struct {
Regex string `protobuf:"bytes,13,opt,name=regex,proto3,oneof"`
}
type SamplingParams_JsonSchema struct {
JsonSchema string `protobuf:"bytes,14,opt,name=json_schema,json=jsonSchema,proto3,oneof"`
}
type SamplingParams_EbnfGrammar struct {
EbnfGrammar string `protobuf:"bytes,15,opt,name=ebnf_grammar,json=ebnfGrammar,proto3,oneof"`
}
type SamplingParams_StructuralTag struct {
StructuralTag string `protobuf:"bytes,16,opt,name=structural_tag,json=structuralTag,proto3,oneof"`
}
func (*SamplingParams_Regex) isSamplingParams_Constraint() {}
func (*SamplingParams_JsonSchema) isSamplingParams_Constraint() {}
func (*SamplingParams_EbnfGrammar) isSamplingParams_Constraint() {}
func (*SamplingParams_StructuralTag) isSamplingParams_Constraint() {}
// Disaggregated serving parameters
type DisaggregatedParams struct {
state protoimpl.MessageState `protogen:"open.v1"`
BootstrapHost string `protobuf:"bytes,1,opt,name=bootstrap_host,json=bootstrapHost,proto3" json:"bootstrap_host,omitempty"`
BootstrapPort int32 `protobuf:"varint,2,opt,name=bootstrap_port,json=bootstrapPort,proto3" json:"bootstrap_port,omitempty"`
BootstrapRoom int32 `protobuf:"varint,3,opt,name=bootstrap_room,json=bootstrapRoom,proto3" json:"bootstrap_room,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *DisaggregatedParams) Reset() {
*x = DisaggregatedParams{}
mi := &file_sglang_scheduler_proto_msgTypes[1]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *DisaggregatedParams) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*DisaggregatedParams) ProtoMessage() {}
func (x *DisaggregatedParams) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[1]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use DisaggregatedParams.ProtoReflect.Descriptor instead.
func (*DisaggregatedParams) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{1}
}
func (x *DisaggregatedParams) GetBootstrapHost() string {
if x != nil {
return x.BootstrapHost
}
return ""
}
func (x *DisaggregatedParams) GetBootstrapPort() int32 {
if x != nil {
return x.BootstrapPort
}
return 0
}
func (x *DisaggregatedParams) GetBootstrapRoom() int32 {
if x != nil {
return x.BootstrapRoom
}
return 0
}
type GenerateRequest struct {
state protoimpl.MessageState `protogen:"open.v1"`
RequestId string `protobuf:"bytes,1,opt,name=request_id,json=requestId,proto3" json:"request_id,omitempty"`
// Input must be tokenized (no raw text)
Tokenized *TokenizedInput `protobuf:"bytes,2,opt,name=tokenized,proto3" json:"tokenized,omitempty"`
// Multimodal inputs
MmInputs *MultimodalInputs `protobuf:"bytes,3,opt,name=mm_inputs,json=mmInputs,proto3" json:"mm_inputs,omitempty"`
// Generation parameters
SamplingParams *SamplingParams `protobuf:"bytes,4,opt,name=sampling_params,json=samplingParams,proto3" json:"sampling_params,omitempty"`
// Return options
ReturnLogprob bool `protobuf:"varint,5,opt,name=return_logprob,json=returnLogprob,proto3" json:"return_logprob,omitempty"`
LogprobStartLen int32 `protobuf:"varint,6,opt,name=logprob_start_len,json=logprobStartLen,proto3" json:"logprob_start_len,omitempty"`
TopLogprobsNum int32 `protobuf:"varint,7,opt,name=top_logprobs_num,json=topLogprobsNum,proto3" json:"top_logprobs_num,omitempty"`
TokenIdsLogprob []uint32 `protobuf:"varint,8,rep,packed,name=token_ids_logprob,json=tokenIdsLogprob,proto3" json:"token_ids_logprob,omitempty"`
ReturnHiddenStates bool `protobuf:"varint,9,opt,name=return_hidden_states,json=returnHiddenStates,proto3" json:"return_hidden_states,omitempty"`
// For disaggregated serving
DisaggregatedParams *DisaggregatedParams `protobuf:"bytes,10,opt,name=disaggregated_params,json=disaggregatedParams,proto3" json:"disaggregated_params,omitempty"`
// Custom logit processor (serialized)
CustomLogitProcessor string `protobuf:"bytes,11,opt,name=custom_logit_processor,json=customLogitProcessor,proto3" json:"custom_logit_processor,omitempty"`
// Request metadata
Timestamp *timestamppb.Timestamp `protobuf:"bytes,12,opt,name=timestamp,proto3" json:"timestamp,omitempty"`
LogMetrics bool `protobuf:"varint,13,opt,name=log_metrics,json=logMetrics,proto3" json:"log_metrics,omitempty"`
// Input embeddings (alternative to text/tokens)
InputEmbeds []float32 `protobuf:"fixed32,14,rep,packed,name=input_embeds,json=inputEmbeds,proto3" json:"input_embeds,omitempty"`
// LoRA adapter ID (if pre-loaded)
LoraId string `protobuf:"bytes,15,opt,name=lora_id,json=loraId,proto3" json:"lora_id,omitempty"`
// Data parallel routing
DataParallelRank int32 `protobuf:"varint,16,opt,name=data_parallel_rank,json=dataParallelRank,proto3" json:"data_parallel_rank,omitempty"`
// Whether client wants streaming response
Stream bool `protobuf:"varint,17,opt,name=stream,proto3" json:"stream,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *GenerateRequest) Reset() {
*x = GenerateRequest{}
mi := &file_sglang_scheduler_proto_msgTypes[2]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *GenerateRequest) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*GenerateRequest) ProtoMessage() {}
func (x *GenerateRequest) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[2]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use GenerateRequest.ProtoReflect.Descriptor instead.
func (*GenerateRequest) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{2}
}
func (x *GenerateRequest) GetRequestId() string {
if x != nil {
return x.RequestId
}
return ""
}
func (x *GenerateRequest) GetTokenized() *TokenizedInput {
if x != nil {
return x.Tokenized
}
return nil
}
func (x *GenerateRequest) GetMmInputs() *MultimodalInputs {
if x != nil {
return x.MmInputs
}
return nil
}
func (x *GenerateRequest) GetSamplingParams() *SamplingParams {
if x != nil {
return x.SamplingParams
}
return nil
}
func (x *GenerateRequest) GetReturnLogprob() bool {
if x != nil {
return x.ReturnLogprob
}
return false
}
func (x *GenerateRequest) GetLogprobStartLen() int32 {
if x != nil {
return x.LogprobStartLen
}
return 0
}
func (x *GenerateRequest) GetTopLogprobsNum() int32 {
if x != nil {
return x.TopLogprobsNum
}
return 0
}
func (x *GenerateRequest) GetTokenIdsLogprob() []uint32 {
if x != nil {
return x.TokenIdsLogprob
}
return nil
}
func (x *GenerateRequest) GetReturnHiddenStates() bool {
if x != nil {
return x.ReturnHiddenStates
}
return false
}
func (x *GenerateRequest) GetDisaggregatedParams() *DisaggregatedParams {
if x != nil {
return x.DisaggregatedParams
}
return nil
}
func (x *GenerateRequest) GetCustomLogitProcessor() string {
if x != nil {
return x.CustomLogitProcessor
}
return ""
}
func (x *GenerateRequest) GetTimestamp() *timestamppb.Timestamp {
if x != nil {
return x.Timestamp
}
return nil
}
func (x *GenerateRequest) GetLogMetrics() bool {
if x != nil {
return x.LogMetrics
}
return false
}
func (x *GenerateRequest) GetInputEmbeds() []float32 {
if x != nil {
return x.InputEmbeds
}
return nil
}
func (x *GenerateRequest) GetLoraId() string {
if x != nil {
return x.LoraId
}
return ""
}
func (x *GenerateRequest) GetDataParallelRank() int32 {
if x != nil {
return x.DataParallelRank
}
return 0
}
func (x *GenerateRequest) GetStream() bool {
if x != nil {
return x.Stream
}
return false
}
type TokenizedInput struct {
state protoimpl.MessageState `protogen:"open.v1"`
OriginalText string `protobuf:"bytes,1,opt,name=original_text,json=originalText,proto3" json:"original_text,omitempty"` // For reference
InputIds []uint32 `protobuf:"varint,2,rep,packed,name=input_ids,json=inputIds,proto3" json:"input_ids,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *TokenizedInput) Reset() {
*x = TokenizedInput{}
mi := &file_sglang_scheduler_proto_msgTypes[3]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *TokenizedInput) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*TokenizedInput) ProtoMessage() {}
func (x *TokenizedInput) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[3]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use TokenizedInput.ProtoReflect.Descriptor instead.
func (*TokenizedInput) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{3}
}
func (x *TokenizedInput) GetOriginalText() string {
if x != nil {
return x.OriginalText
}
return ""
}
func (x *TokenizedInput) GetInputIds() []uint32 {
if x != nil {
return x.InputIds
}
return nil
}
type MultimodalInputs struct {
state protoimpl.MessageState `protogen:"open.v1"`
// Simplified multimodal handling - actual data processed by tokenizer
ImageUrls []string `protobuf:"bytes,1,rep,name=image_urls,json=imageUrls,proto3" json:"image_urls,omitempty"`
VideoUrls []string `protobuf:"bytes,2,rep,name=video_urls,json=videoUrls,proto3" json:"video_urls,omitempty"`
AudioUrls []string `protobuf:"bytes,3,rep,name=audio_urls,json=audioUrls,proto3" json:"audio_urls,omitempty"`
// Pre-processed multimodal features (if available)
ProcessedFeatures *structpb.Struct `protobuf:"bytes,4,opt,name=processed_features,json=processedFeatures,proto3" json:"processed_features,omitempty"`
// Raw data for direct processing
ImageData [][]byte `protobuf:"bytes,5,rep,name=image_data,json=imageData,proto3" json:"image_data,omitempty"`
VideoData [][]byte `protobuf:"bytes,6,rep,name=video_data,json=videoData,proto3" json:"video_data,omitempty"`
AudioData [][]byte `protobuf:"bytes,7,rep,name=audio_data,json=audioData,proto3" json:"audio_data,omitempty"`
// Modality metadata
Modalities []string `protobuf:"bytes,8,rep,name=modalities,proto3" json:"modalities,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *MultimodalInputs) Reset() {
*x = MultimodalInputs{}
mi := &file_sglang_scheduler_proto_msgTypes[4]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *MultimodalInputs) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*MultimodalInputs) ProtoMessage() {}
func (x *MultimodalInputs) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[4]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use MultimodalInputs.ProtoReflect.Descriptor instead.
func (*MultimodalInputs) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{4}
}
func (x *MultimodalInputs) GetImageUrls() []string {
if x != nil {
return x.ImageUrls
}
return nil
}
func (x *MultimodalInputs) GetVideoUrls() []string {
if x != nil {
return x.VideoUrls
}
return nil
}
func (x *MultimodalInputs) GetAudioUrls() []string {
if x != nil {
return x.AudioUrls
}
return nil
}
func (x *MultimodalInputs) GetProcessedFeatures() *structpb.Struct {
if x != nil {
return x.ProcessedFeatures
}
return nil
}
func (x *MultimodalInputs) GetImageData() [][]byte {
if x != nil {
return x.ImageData
}
return nil
}
func (x *MultimodalInputs) GetVideoData() [][]byte {
if x != nil {
return x.VideoData
}
return nil
}
func (x *MultimodalInputs) GetAudioData() [][]byte {
if x != nil {
return x.AudioData
}
return nil
}
func (x *MultimodalInputs) GetModalities() []string {
if x != nil {
return x.Modalities
}
return nil
}
type GenerateResponse struct {
state protoimpl.MessageState `protogen:"open.v1"`
RequestId string `protobuf:"bytes,1,opt,name=request_id,json=requestId,proto3" json:"request_id,omitempty"`
// Response type
//
// Types that are valid to be assigned to Response:
//
// *GenerateResponse_Chunk
// *GenerateResponse_Complete
// *GenerateResponse_Error
Response isGenerateResponse_Response `protobuf_oneof:"response"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *GenerateResponse) Reset() {
*x = GenerateResponse{}
mi := &file_sglang_scheduler_proto_msgTypes[5]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *GenerateResponse) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*GenerateResponse) ProtoMessage() {}
func (x *GenerateResponse) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[5]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use GenerateResponse.ProtoReflect.Descriptor instead.
func (*GenerateResponse) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{5}
}
func (x *GenerateResponse) GetRequestId() string {
if x != nil {
return x.RequestId
}
return ""
}
func (x *GenerateResponse) GetResponse() isGenerateResponse_Response {
if x != nil {
return x.Response
}
return nil
}
func (x *GenerateResponse) GetChunk() *GenerateStreamChunk {
if x != nil {
if x, ok := x.Response.(*GenerateResponse_Chunk); ok {
return x.Chunk
}
}
return nil
}
func (x *GenerateResponse) GetComplete() *GenerateComplete {
if x != nil {
if x, ok := x.Response.(*GenerateResponse_Complete); ok {
return x.Complete
}
}
return nil
}
func (x *GenerateResponse) GetError() *GenerateError {
if x != nil {
if x, ok := x.Response.(*GenerateResponse_Error); ok {
return x.Error
}
}
return nil
}
type isGenerateResponse_Response interface {
isGenerateResponse_Response()
}
type GenerateResponse_Chunk struct {
Chunk *GenerateStreamChunk `protobuf:"bytes,2,opt,name=chunk,proto3,oneof"`
}
type GenerateResponse_Complete struct {
Complete *GenerateComplete `protobuf:"bytes,3,opt,name=complete,proto3,oneof"`
}
type GenerateResponse_Error struct {
Error *GenerateError `protobuf:"bytes,4,opt,name=error,proto3,oneof"`
}
func (*GenerateResponse_Chunk) isGenerateResponse_Response() {}
func (*GenerateResponse_Complete) isGenerateResponse_Response() {}
func (*GenerateResponse_Error) isGenerateResponse_Response() {}
type GenerateStreamChunk struct {
state protoimpl.MessageState `protogen:"open.v1"`
// Generated tokens (incremental chunk)
TokenIds []uint32 `protobuf:"varint,1,rep,packed,name=token_ids,json=tokenIds,proto3" json:"token_ids,omitempty"`
// Cumulative counts
PromptTokens int32 `protobuf:"varint,2,opt,name=prompt_tokens,json=promptTokens,proto3" json:"prompt_tokens,omitempty"`
CompletionTokens int32 `protobuf:"varint,3,opt,name=completion_tokens,json=completionTokens,proto3" json:"completion_tokens,omitempty"`
CachedTokens int32 `protobuf:"varint,4,opt,name=cached_tokens,json=cachedTokens,proto3" json:"cached_tokens,omitempty"`
// Output logprobs (if requested) - incremental for streaming
OutputLogprobs *OutputLogProbs `protobuf:"bytes,5,opt,name=output_logprobs,json=outputLogprobs,proto3" json:"output_logprobs,omitempty"`
// Hidden states (if requested)
HiddenStates []float32 `protobuf:"fixed32,6,rep,packed,name=hidden_states,json=hiddenStates,proto3" json:"hidden_states,omitempty"`
// Input logprobs (if requested) - only in first chunk
InputLogprobs *InputLogProbs `protobuf:"bytes,7,opt,name=input_logprobs,json=inputLogprobs,proto3" json:"input_logprobs,omitempty"`
// Index for ordering when n>1 (for parallel request multiplexing)
Index uint32 `protobuf:"varint,8,opt,name=index,proto3" json:"index,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *GenerateStreamChunk) Reset() {
*x = GenerateStreamChunk{}
mi := &file_sglang_scheduler_proto_msgTypes[6]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *GenerateStreamChunk) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*GenerateStreamChunk) ProtoMessage() {}
func (x *GenerateStreamChunk) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[6]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use GenerateStreamChunk.ProtoReflect.Descriptor instead.
func (*GenerateStreamChunk) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{6}
}
func (x *GenerateStreamChunk) GetTokenIds() []uint32 {
if x != nil {
return x.TokenIds
}
return nil
}
func (x *GenerateStreamChunk) GetPromptTokens() int32 {
if x != nil {
return x.PromptTokens
}
return 0
}
func (x *GenerateStreamChunk) GetCompletionTokens() int32 {
if x != nil {
return x.CompletionTokens
}
return 0
}
func (x *GenerateStreamChunk) GetCachedTokens() int32 {
if x != nil {
return x.CachedTokens
}
return 0
}
func (x *GenerateStreamChunk) GetOutputLogprobs() *OutputLogProbs {
if x != nil {
return x.OutputLogprobs
}
return nil
}
func (x *GenerateStreamChunk) GetHiddenStates() []float32 {
if x != nil {
return x.HiddenStates
}
return nil
}
func (x *GenerateStreamChunk) GetInputLogprobs() *InputLogProbs {
if x != nil {
return x.InputLogprobs
}
return nil
}
func (x *GenerateStreamChunk) GetIndex() uint32 {
if x != nil {
return x.Index
}
return 0
}
type GenerateComplete struct {
state protoimpl.MessageState `protogen:"open.v1"`
// Final output
OutputIds []uint32 `protobuf:"varint,1,rep,packed,name=output_ids,json=outputIds,proto3" json:"output_ids,omitempty"`
// Finish reason as OpenAI-compatible string ("stop", "length", "abort")
FinishReason string `protobuf:"bytes,2,opt,name=finish_reason,json=finishReason,proto3" json:"finish_reason,omitempty"`
// Token usage counts
PromptTokens int32 `protobuf:"varint,3,opt,name=prompt_tokens,json=promptTokens,proto3" json:"prompt_tokens,omitempty"`
CompletionTokens int32 `protobuf:"varint,4,opt,name=completion_tokens,json=completionTokens,proto3" json:"completion_tokens,omitempty"`
CachedTokens int32 `protobuf:"varint,5,opt,name=cached_tokens,json=cachedTokens,proto3" json:"cached_tokens,omitempty"`
// Output logprobs if requested (cumulative)
OutputLogprobs *OutputLogProbs `protobuf:"bytes,6,opt,name=output_logprobs,json=outputLogprobs,proto3" json:"output_logprobs,omitempty"`
// All hidden states if requested
AllHiddenStates []*HiddenStates `protobuf:"bytes,7,rep,name=all_hidden_states,json=allHiddenStates,proto3" json:"all_hidden_states,omitempty"`
// Matched stop information (for stop sequences)
//
// Types that are valid to be assigned to MatchedStop:
//
// *GenerateComplete_MatchedTokenId
// *GenerateComplete_MatchedStopStr
MatchedStop isGenerateComplete_MatchedStop `protobuf_oneof:"matched_stop"`
// Input logprobs if requested (for prompt tokens)
InputLogprobs *InputLogProbs `protobuf:"bytes,10,opt,name=input_logprobs,json=inputLogprobs,proto3" json:"input_logprobs,omitempty"`
// Index for ordering when n>1 (for parallel request multiplexing)
Index uint32 `protobuf:"varint,11,opt,name=index,proto3" json:"index,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *GenerateComplete) Reset() {
*x = GenerateComplete{}
mi := &file_sglang_scheduler_proto_msgTypes[7]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *GenerateComplete) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*GenerateComplete) ProtoMessage() {}
func (x *GenerateComplete) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[7]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use GenerateComplete.ProtoReflect.Descriptor instead.
func (*GenerateComplete) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{7}
}
func (x *GenerateComplete) GetOutputIds() []uint32 {
if x != nil {
return x.OutputIds
}
return nil
}
func (x *GenerateComplete) GetFinishReason() string {
if x != nil {
return x.FinishReason
}
return ""
}
func (x *GenerateComplete) GetPromptTokens() int32 {
if x != nil {
return x.PromptTokens
}
return 0
}
func (x *GenerateComplete) GetCompletionTokens() int32 {
if x != nil {
return x.CompletionTokens
}
return 0
}
func (x *GenerateComplete) GetCachedTokens() int32 {
if x != nil {
return x.CachedTokens
}
return 0
}
func (x *GenerateComplete) GetOutputLogprobs() *OutputLogProbs {
if x != nil {
return x.OutputLogprobs
}
return nil
}
func (x *GenerateComplete) GetAllHiddenStates() []*HiddenStates {
if x != nil {
return x.AllHiddenStates
}
return nil
}
func (x *GenerateComplete) GetMatchedStop() isGenerateComplete_MatchedStop {
if x != nil {
return x.MatchedStop
}
return nil
}
func (x *GenerateComplete) GetMatchedTokenId() uint32 {
if x != nil {
if x, ok := x.MatchedStop.(*GenerateComplete_MatchedTokenId); ok {
return x.MatchedTokenId
}
}
return 0
}
func (x *GenerateComplete) GetMatchedStopStr() string {
if x != nil {
if x, ok := x.MatchedStop.(*GenerateComplete_MatchedStopStr); ok {
return x.MatchedStopStr
}
}
return ""
}
func (x *GenerateComplete) GetInputLogprobs() *InputLogProbs {
if x != nil {
return x.InputLogprobs
}
return nil
}
func (x *GenerateComplete) GetIndex() uint32 {
if x != nil {
return x.Index
}
return 0
}
type isGenerateComplete_MatchedStop interface {
isGenerateComplete_MatchedStop()
}
type GenerateComplete_MatchedTokenId struct {
MatchedTokenId uint32 `protobuf:"varint,8,opt,name=matched_token_id,json=matchedTokenId,proto3,oneof"`
}
type GenerateComplete_MatchedStopStr struct {
MatchedStopStr string `protobuf:"bytes,9,opt,name=matched_stop_str,json=matchedStopStr,proto3,oneof"`
}
func (*GenerateComplete_MatchedTokenId) isGenerateComplete_MatchedStop() {}
func (*GenerateComplete_MatchedStopStr) isGenerateComplete_MatchedStop() {}
type GenerateError struct {
state protoimpl.MessageState `protogen:"open.v1"`
Message string `protobuf:"bytes,1,opt,name=message,proto3" json:"message,omitempty"`
HttpStatusCode string `protobuf:"bytes,2,opt,name=http_status_code,json=httpStatusCode,proto3" json:"http_status_code,omitempty"`
Details string `protobuf:"bytes,3,opt,name=details,proto3" json:"details,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *GenerateError) Reset() {
*x = GenerateError{}
mi := &file_sglang_scheduler_proto_msgTypes[8]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *GenerateError) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*GenerateError) ProtoMessage() {}
func (x *GenerateError) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[8]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use GenerateError.ProtoReflect.Descriptor instead.
func (*GenerateError) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{8}
}
func (x *GenerateError) GetMessage() string {
if x != nil {
return x.Message
}
return ""
}
func (x *GenerateError) GetHttpStatusCode() string {
if x != nil {
return x.HttpStatusCode
}
return ""
}
func (x *GenerateError) GetDetails() string {
if x != nil {
return x.Details
}
return ""
}
// Output logprobs - all values are present (no None)
type OutputLogProbs struct {
state protoimpl.MessageState `protogen:"open.v1"`
TokenLogprobs []float32 `protobuf:"fixed32,1,rep,packed,name=token_logprobs,json=tokenLogprobs,proto3" json:"token_logprobs,omitempty"`
TokenIds []int32 `protobuf:"varint,2,rep,packed,name=token_ids,json=tokenIds,proto3" json:"token_ids,omitempty"`
// Top logprobs at each position
TopLogprobs []*TopLogProbs `protobuf:"bytes,3,rep,name=top_logprobs,json=topLogprobs,proto3" json:"top_logprobs,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *OutputLogProbs) Reset() {
*x = OutputLogProbs{}
mi := &file_sglang_scheduler_proto_msgTypes[9]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *OutputLogProbs) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*OutputLogProbs) ProtoMessage() {}
func (x *OutputLogProbs) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[9]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use OutputLogProbs.ProtoReflect.Descriptor instead.
func (*OutputLogProbs) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{9}
}
func (x *OutputLogProbs) GetTokenLogprobs() []float32 {
if x != nil {
return x.TokenLogprobs
}
return nil
}
func (x *OutputLogProbs) GetTokenIds() []int32 {
if x != nil {
return x.TokenIds
}
return nil
}
func (x *OutputLogProbs) GetTopLogprobs() []*TopLogProbs {
if x != nil {
return x.TopLogprobs
}
return nil
}
// Input logprobs - first token has no logprob (None)
type InputLogProbs struct {
state protoimpl.MessageState `protogen:"open.v1"`
TokenLogprobs []*InputTokenLogProb `protobuf:"bytes,1,rep,name=token_logprobs,json=tokenLogprobs,proto3" json:"token_logprobs,omitempty"`
TokenIds []int32 `protobuf:"varint,2,rep,packed,name=token_ids,json=tokenIds,proto3" json:"token_ids,omitempty"`
// Top logprobs at each position
TopLogprobs []*TopLogProbs `protobuf:"bytes,3,rep,name=top_logprobs,json=topLogprobs,proto3" json:"top_logprobs,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *InputLogProbs) Reset() {
*x = InputLogProbs{}
mi := &file_sglang_scheduler_proto_msgTypes[10]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *InputLogProbs) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*InputLogProbs) ProtoMessage() {}
func (x *InputLogProbs) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[10]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use InputLogProbs.ProtoReflect.Descriptor instead.
func (*InputLogProbs) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{10}
}
func (x *InputLogProbs) GetTokenLogprobs() []*InputTokenLogProb {
if x != nil {
return x.TokenLogprobs
}
return nil
}
func (x *InputLogProbs) GetTokenIds() []int32 {
if x != nil {
return x.TokenIds
}
return nil
}
func (x *InputLogProbs) GetTopLogprobs() []*TopLogProbs {
if x != nil {
return x.TopLogprobs
}
return nil
}
// Wrapper to represent optional logprob (first input token has no logprob)
type InputTokenLogProb struct {
state protoimpl.MessageState `protogen:"open.v1"`
Value *float32 `protobuf:"fixed32,1,opt,name=value,proto3,oneof" json:"value,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *InputTokenLogProb) Reset() {
*x = InputTokenLogProb{}
mi := &file_sglang_scheduler_proto_msgTypes[11]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *InputTokenLogProb) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*InputTokenLogProb) ProtoMessage() {}
func (x *InputTokenLogProb) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[11]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use InputTokenLogProb.ProtoReflect.Descriptor instead.
func (*InputTokenLogProb) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{11}
}
func (x *InputTokenLogProb) GetValue() float32 {
if x != nil && x.Value != nil {
return *x.Value
}
return 0
}
type TopLogProbs struct {
state protoimpl.MessageState `protogen:"open.v1"`
Values []float32 `protobuf:"fixed32,1,rep,packed,name=values,proto3" json:"values,omitempty"`
TokenIds []int32 `protobuf:"varint,2,rep,packed,name=token_ids,json=tokenIds,proto3" json:"token_ids,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *TopLogProbs) Reset() {
*x = TopLogProbs{}
mi := &file_sglang_scheduler_proto_msgTypes[12]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *TopLogProbs) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*TopLogProbs) ProtoMessage() {}
func (x *TopLogProbs) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[12]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use TopLogProbs.ProtoReflect.Descriptor instead.
func (*TopLogProbs) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{12}
}
func (x *TopLogProbs) GetValues() []float32 {
if x != nil {
return x.Values
}
return nil
}
func (x *TopLogProbs) GetTokenIds() []int32 {
if x != nil {
return x.TokenIds
}
return nil
}
type HiddenStates struct {
state protoimpl.MessageState `protogen:"open.v1"`
Values []float32 `protobuf:"fixed32,1,rep,packed,name=values,proto3" json:"values,omitempty"`
Layer int32 `protobuf:"varint,2,opt,name=layer,proto3" json:"layer,omitempty"`
Position int32 `protobuf:"varint,3,opt,name=position,proto3" json:"position,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *HiddenStates) Reset() {
*x = HiddenStates{}
mi := &file_sglang_scheduler_proto_msgTypes[13]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *HiddenStates) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*HiddenStates) ProtoMessage() {}
func (x *HiddenStates) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[13]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use HiddenStates.ProtoReflect.Descriptor instead.
func (*HiddenStates) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{13}
}
func (x *HiddenStates) GetValues() []float32 {
if x != nil {
return x.Values
}
return nil
}
func (x *HiddenStates) GetLayer() int32 {
if x != nil {
return x.Layer
}
return 0
}
func (x *HiddenStates) GetPosition() int32 {
if x != nil {
return x.Position
}
return 0
}
type EmbedRequest struct {
state protoimpl.MessageState `protogen:"open.v1"`
RequestId string `protobuf:"bytes,1,opt,name=request_id,json=requestId,proto3" json:"request_id,omitempty"`
// Input must be tokenized (no raw text)
Tokenized *TokenizedInput `protobuf:"bytes,2,opt,name=tokenized,proto3" json:"tokenized,omitempty"`
// Multimodal inputs
MmInputs *MultimodalInputs `protobuf:"bytes,4,opt,name=mm_inputs,json=mmInputs,proto3" json:"mm_inputs,omitempty"`
// Dummy sampling params for compatibility
// EmbedRequest doesn't use sampling_params
SamplingParams *SamplingParams `protobuf:"bytes,5,opt,name=sampling_params,json=samplingParams,proto3" json:"sampling_params,omitempty"`
LogMetrics bool `protobuf:"varint,6,opt,name=log_metrics,json=logMetrics,proto3" json:"log_metrics,omitempty"`
// Token type IDs for models that require them
TokenTypeIds []int32 `protobuf:"varint,7,rep,packed,name=token_type_ids,json=tokenTypeIds,proto3" json:"token_type_ids,omitempty"`
// Data parallel routing
DataParallelRank int32 `protobuf:"varint,8,opt,name=data_parallel_rank,json=dataParallelRank,proto3" json:"data_parallel_rank,omitempty"`
// For cross-encoder requests
IsCrossEncoder bool `protobuf:"varint,9,opt,name=is_cross_encoder,json=isCrossEncoder,proto3" json:"is_cross_encoder,omitempty"`
Texts []string `protobuf:"bytes,10,rep,name=texts,proto3" json:"texts,omitempty"` // For cross-encoder batch
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *EmbedRequest) Reset() {
*x = EmbedRequest{}
mi := &file_sglang_scheduler_proto_msgTypes[14]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *EmbedRequest) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*EmbedRequest) ProtoMessage() {}
func (x *EmbedRequest) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[14]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use EmbedRequest.ProtoReflect.Descriptor instead.
func (*EmbedRequest) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{14}
}
func (x *EmbedRequest) GetRequestId() string {
if x != nil {
return x.RequestId
}
return ""
}
func (x *EmbedRequest) GetTokenized() *TokenizedInput {
if x != nil {
return x.Tokenized
}
return nil
}
func (x *EmbedRequest) GetMmInputs() *MultimodalInputs {
if x != nil {
return x.MmInputs
}
return nil
}
func (x *EmbedRequest) GetSamplingParams() *SamplingParams {
if x != nil {
return x.SamplingParams
}
return nil
}
func (x *EmbedRequest) GetLogMetrics() bool {
if x != nil {
return x.LogMetrics
}
return false
}
func (x *EmbedRequest) GetTokenTypeIds() []int32 {
if x != nil {
return x.TokenTypeIds
}
return nil
}
func (x *EmbedRequest) GetDataParallelRank() int32 {
if x != nil {
return x.DataParallelRank
}
return 0
}
func (x *EmbedRequest) GetIsCrossEncoder() bool {
if x != nil {
return x.IsCrossEncoder
}
return false
}
func (x *EmbedRequest) GetTexts() []string {
if x != nil {
return x.Texts
}
return nil
}
type EmbedResponse struct {
state protoimpl.MessageState `protogen:"open.v1"`
RequestId string `protobuf:"bytes,1,opt,name=request_id,json=requestId,proto3" json:"request_id,omitempty"`
// Types that are valid to be assigned to Response:
//
// *EmbedResponse_Complete
// *EmbedResponse_Error
Response isEmbedResponse_Response `protobuf_oneof:"response"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *EmbedResponse) Reset() {
*x = EmbedResponse{}
mi := &file_sglang_scheduler_proto_msgTypes[15]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *EmbedResponse) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*EmbedResponse) ProtoMessage() {}
func (x *EmbedResponse) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[15]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use EmbedResponse.ProtoReflect.Descriptor instead.
func (*EmbedResponse) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{15}
}
func (x *EmbedResponse) GetRequestId() string {
if x != nil {
return x.RequestId
}
return ""
}
func (x *EmbedResponse) GetResponse() isEmbedResponse_Response {
if x != nil {
return x.Response
}
return nil
}
func (x *EmbedResponse) GetComplete() *EmbedComplete {
if x != nil {
if x, ok := x.Response.(*EmbedResponse_Complete); ok {
return x.Complete
}
}
return nil
}
func (x *EmbedResponse) GetError() *EmbedError {
if x != nil {
if x, ok := x.Response.(*EmbedResponse_Error); ok {
return x.Error
}
}
return nil
}
type isEmbedResponse_Response interface {
isEmbedResponse_Response()
}
type EmbedResponse_Complete struct {
Complete *EmbedComplete `protobuf:"bytes,2,opt,name=complete,proto3,oneof"`
}
type EmbedResponse_Error struct {
Error *EmbedError `protobuf:"bytes,3,opt,name=error,proto3,oneof"`
}
func (*EmbedResponse_Complete) isEmbedResponse_Response() {}
func (*EmbedResponse_Error) isEmbedResponse_Response() {}
type EmbedComplete struct {
state protoimpl.MessageState `protogen:"open.v1"`
Embedding []float32 `protobuf:"fixed32,1,rep,packed,name=embedding,proto3" json:"embedding,omitempty"`
PromptTokens int32 `protobuf:"varint,2,opt,name=prompt_tokens,json=promptTokens,proto3" json:"prompt_tokens,omitempty"`
CachedTokens int32 `protobuf:"varint,3,opt,name=cached_tokens,json=cachedTokens,proto3" json:"cached_tokens,omitempty"`
// Additional metadata
EmbeddingDim int32 `protobuf:"varint,4,opt,name=embedding_dim,json=embeddingDim,proto3" json:"embedding_dim,omitempty"`
// For batch embeddings
BatchEmbeddings []*Embedding `protobuf:"bytes,5,rep,name=batch_embeddings,json=batchEmbeddings,proto3" json:"batch_embeddings,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *EmbedComplete) Reset() {
*x = EmbedComplete{}
mi := &file_sglang_scheduler_proto_msgTypes[16]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *EmbedComplete) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*EmbedComplete) ProtoMessage() {}
func (x *EmbedComplete) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[16]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use EmbedComplete.ProtoReflect.Descriptor instead.
func (*EmbedComplete) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{16}
}
func (x *EmbedComplete) GetEmbedding() []float32 {
if x != nil {
return x.Embedding
}
return nil
}
func (x *EmbedComplete) GetPromptTokens() int32 {
if x != nil {
return x.PromptTokens
}
return 0
}
func (x *EmbedComplete) GetCachedTokens() int32 {
if x != nil {
return x.CachedTokens
}
return 0
}
func (x *EmbedComplete) GetEmbeddingDim() int32 {
if x != nil {
return x.EmbeddingDim
}
return 0
}
func (x *EmbedComplete) GetBatchEmbeddings() []*Embedding {
if x != nil {
return x.BatchEmbeddings
}
return nil
}
type Embedding struct {
state protoimpl.MessageState `protogen:"open.v1"`
Values []float32 `protobuf:"fixed32,1,rep,packed,name=values,proto3" json:"values,omitempty"`
Index int32 `protobuf:"varint,2,opt,name=index,proto3" json:"index,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *Embedding) Reset() {
*x = Embedding{}
mi := &file_sglang_scheduler_proto_msgTypes[17]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *Embedding) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*Embedding) ProtoMessage() {}
func (x *Embedding) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[17]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use Embedding.ProtoReflect.Descriptor instead.
func (*Embedding) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{17}
}
func (x *Embedding) GetValues() []float32 {
if x != nil {
return x.Values
}
return nil
}
func (x *Embedding) GetIndex() int32 {
if x != nil {
return x.Index
}
return 0
}
type EmbedError struct {
state protoimpl.MessageState `protogen:"open.v1"`
Message string `protobuf:"bytes,1,opt,name=message,proto3" json:"message,omitempty"`
Code string `protobuf:"bytes,2,opt,name=code,proto3" json:"code,omitempty"`
Details string `protobuf:"bytes,3,opt,name=details,proto3" json:"details,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *EmbedError) Reset() {
*x = EmbedError{}
mi := &file_sglang_scheduler_proto_msgTypes[18]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *EmbedError) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*EmbedError) ProtoMessage() {}
func (x *EmbedError) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[18]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use EmbedError.ProtoReflect.Descriptor instead.
func (*EmbedError) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{18}
}
func (x *EmbedError) GetMessage() string {
if x != nil {
return x.Message
}
return ""
}
func (x *EmbedError) GetCode() string {
if x != nil {
return x.Code
}
return ""
}
func (x *EmbedError) GetDetails() string {
if x != nil {
return x.Details
}
return ""
}
type HealthCheckRequest struct {
state protoimpl.MessageState `protogen:"open.v1"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *HealthCheckRequest) Reset() {
*x = HealthCheckRequest{}
mi := &file_sglang_scheduler_proto_msgTypes[19]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *HealthCheckRequest) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*HealthCheckRequest) ProtoMessage() {}
func (x *HealthCheckRequest) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[19]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use HealthCheckRequest.ProtoReflect.Descriptor instead.
func (*HealthCheckRequest) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{19}
}
type HealthCheckResponse struct {
state protoimpl.MessageState `protogen:"open.v1"`
Healthy bool `protobuf:"varint,1,opt,name=healthy,proto3" json:"healthy,omitempty"`
Message string `protobuf:"bytes,2,opt,name=message,proto3" json:"message,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *HealthCheckResponse) Reset() {
*x = HealthCheckResponse{}
mi := &file_sglang_scheduler_proto_msgTypes[20]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *HealthCheckResponse) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*HealthCheckResponse) ProtoMessage() {}
func (x *HealthCheckResponse) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[20]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use HealthCheckResponse.ProtoReflect.Descriptor instead.
func (*HealthCheckResponse) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{20}
}
func (x *HealthCheckResponse) GetHealthy() bool {
if x != nil {
return x.Healthy
}
return false
}
func (x *HealthCheckResponse) GetMessage() string {
if x != nil {
return x.Message
}
return ""
}
type AbortRequest struct {
state protoimpl.MessageState `protogen:"open.v1"`
RequestId string `protobuf:"bytes,1,opt,name=request_id,json=requestId,proto3" json:"request_id,omitempty"`
Reason string `protobuf:"bytes,2,opt,name=reason,proto3" json:"reason,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *AbortRequest) Reset() {
*x = AbortRequest{}
mi := &file_sglang_scheduler_proto_msgTypes[21]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *AbortRequest) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*AbortRequest) ProtoMessage() {}
func (x *AbortRequest) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[21]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use AbortRequest.ProtoReflect.Descriptor instead.
func (*AbortRequest) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{21}
}
func (x *AbortRequest) GetRequestId() string {
if x != nil {
return x.RequestId
}
return ""
}
func (x *AbortRequest) GetReason() string {
if x != nil {
return x.Reason
}
return ""
}
type AbortResponse struct {
state protoimpl.MessageState `protogen:"open.v1"`
Success bool `protobuf:"varint,1,opt,name=success,proto3" json:"success,omitempty"`
Message string `protobuf:"bytes,2,opt,name=message,proto3" json:"message,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *AbortResponse) Reset() {
*x = AbortResponse{}
mi := &file_sglang_scheduler_proto_msgTypes[22]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *AbortResponse) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*AbortResponse) ProtoMessage() {}
func (x *AbortResponse) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[22]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use AbortResponse.ProtoReflect.Descriptor instead.
func (*AbortResponse) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{22}
}
func (x *AbortResponse) GetSuccess() bool {
if x != nil {
return x.Success
}
return false
}
func (x *AbortResponse) GetMessage() string {
if x != nil {
return x.Message
}
return ""
}
// Load LoRA adapter
type LoadLoRARequest struct {
state protoimpl.MessageState `protogen:"open.v1"`
AdapterId string `protobuf:"bytes,1,opt,name=adapter_id,json=adapterId,proto3" json:"adapter_id,omitempty"`
AdapterPath string `protobuf:"bytes,2,opt,name=adapter_path,json=adapterPath,proto3" json:"adapter_path,omitempty"`
Rank int32 `protobuf:"varint,3,opt,name=rank,proto3" json:"rank,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *LoadLoRARequest) Reset() {
*x = LoadLoRARequest{}
mi := &file_sglang_scheduler_proto_msgTypes[23]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *LoadLoRARequest) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*LoadLoRARequest) ProtoMessage() {}
func (x *LoadLoRARequest) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[23]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use LoadLoRARequest.ProtoReflect.Descriptor instead.
func (*LoadLoRARequest) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{23}
}
func (x *LoadLoRARequest) GetAdapterId() string {
if x != nil {
return x.AdapterId
}
return ""
}
func (x *LoadLoRARequest) GetAdapterPath() string {
if x != nil {
return x.AdapterPath
}
return ""
}
func (x *LoadLoRARequest) GetRank() int32 {
if x != nil {
return x.Rank
}
return 0
}
type LoadLoRAResponse struct {
state protoimpl.MessageState `protogen:"open.v1"`
Success bool `protobuf:"varint,1,opt,name=success,proto3" json:"success,omitempty"`
AdapterId string `protobuf:"bytes,2,opt,name=adapter_id,json=adapterId,proto3" json:"adapter_id,omitempty"`
Message string `protobuf:"bytes,3,opt,name=message,proto3" json:"message,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *LoadLoRAResponse) Reset() {
*x = LoadLoRAResponse{}
mi := &file_sglang_scheduler_proto_msgTypes[24]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *LoadLoRAResponse) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*LoadLoRAResponse) ProtoMessage() {}
func (x *LoadLoRAResponse) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[24]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use LoadLoRAResponse.ProtoReflect.Descriptor instead.
func (*LoadLoRAResponse) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{24}
}
func (x *LoadLoRAResponse) GetSuccess() bool {
if x != nil {
return x.Success
}
return false
}
func (x *LoadLoRAResponse) GetAdapterId() string {
if x != nil {
return x.AdapterId
}
return ""
}
func (x *LoadLoRAResponse) GetMessage() string {
if x != nil {
return x.Message
}
return ""
}
// Unload LoRA adapter
type UnloadLoRARequest struct {
state protoimpl.MessageState `protogen:"open.v1"`
AdapterId string `protobuf:"bytes,1,opt,name=adapter_id,json=adapterId,proto3" json:"adapter_id,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *UnloadLoRARequest) Reset() {
*x = UnloadLoRARequest{}
mi := &file_sglang_scheduler_proto_msgTypes[25]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *UnloadLoRARequest) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*UnloadLoRARequest) ProtoMessage() {}
func (x *UnloadLoRARequest) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[25]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use UnloadLoRARequest.ProtoReflect.Descriptor instead.
func (*UnloadLoRARequest) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{25}
}
func (x *UnloadLoRARequest) GetAdapterId() string {
if x != nil {
return x.AdapterId
}
return ""
}
type UnloadLoRAResponse struct {
state protoimpl.MessageState `protogen:"open.v1"`
Success bool `protobuf:"varint,1,opt,name=success,proto3" json:"success,omitempty"`
Message string `protobuf:"bytes,2,opt,name=message,proto3" json:"message,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *UnloadLoRAResponse) Reset() {
*x = UnloadLoRAResponse{}
mi := &file_sglang_scheduler_proto_msgTypes[26]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *UnloadLoRAResponse) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*UnloadLoRAResponse) ProtoMessage() {}
func (x *UnloadLoRAResponse) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[26]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use UnloadLoRAResponse.ProtoReflect.Descriptor instead.
func (*UnloadLoRAResponse) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{26}
}
func (x *UnloadLoRAResponse) GetSuccess() bool {
if x != nil {
return x.Success
}
return false
}
func (x *UnloadLoRAResponse) GetMessage() string {
if x != nil {
return x.Message
}
return ""
}
// Update weights
type UpdateWeightsRequest struct {
state protoimpl.MessageState `protogen:"open.v1"`
// Types that are valid to be assigned to Source:
//
// *UpdateWeightsRequest_DiskPath
// *UpdateWeightsRequest_TensorData
// *UpdateWeightsRequest_RemoteUrl
Source isUpdateWeightsRequest_Source `protobuf_oneof:"source"`
WeightName string `protobuf:"bytes,4,opt,name=weight_name,json=weightName,proto3" json:"weight_name,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *UpdateWeightsRequest) Reset() {
*x = UpdateWeightsRequest{}
mi := &file_sglang_scheduler_proto_msgTypes[27]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *UpdateWeightsRequest) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*UpdateWeightsRequest) ProtoMessage() {}
func (x *UpdateWeightsRequest) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[27]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use UpdateWeightsRequest.ProtoReflect.Descriptor instead.
func (*UpdateWeightsRequest) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{27}
}
func (x *UpdateWeightsRequest) GetSource() isUpdateWeightsRequest_Source {
if x != nil {
return x.Source
}
return nil
}
func (x *UpdateWeightsRequest) GetDiskPath() string {
if x != nil {
if x, ok := x.Source.(*UpdateWeightsRequest_DiskPath); ok {
return x.DiskPath
}
}
return ""
}
func (x *UpdateWeightsRequest) GetTensorData() []byte {
if x != nil {
if x, ok := x.Source.(*UpdateWeightsRequest_TensorData); ok {
return x.TensorData
}
}
return nil
}
func (x *UpdateWeightsRequest) GetRemoteUrl() string {
if x != nil {
if x, ok := x.Source.(*UpdateWeightsRequest_RemoteUrl); ok {
return x.RemoteUrl
}
}
return ""
}
func (x *UpdateWeightsRequest) GetWeightName() string {
if x != nil {
return x.WeightName
}
return ""
}
type isUpdateWeightsRequest_Source interface {
isUpdateWeightsRequest_Source()
}
type UpdateWeightsRequest_DiskPath struct {
DiskPath string `protobuf:"bytes,1,opt,name=disk_path,json=diskPath,proto3,oneof"`
}
type UpdateWeightsRequest_TensorData struct {
TensorData []byte `protobuf:"bytes,2,opt,name=tensor_data,json=tensorData,proto3,oneof"`
}
type UpdateWeightsRequest_RemoteUrl struct {
RemoteUrl string `protobuf:"bytes,3,opt,name=remote_url,json=remoteUrl,proto3,oneof"`
}
func (*UpdateWeightsRequest_DiskPath) isUpdateWeightsRequest_Source() {}
func (*UpdateWeightsRequest_TensorData) isUpdateWeightsRequest_Source() {}
func (*UpdateWeightsRequest_RemoteUrl) isUpdateWeightsRequest_Source() {}
type UpdateWeightsResponse struct {
state protoimpl.MessageState `protogen:"open.v1"`
Success bool `protobuf:"varint,1,opt,name=success,proto3" json:"success,omitempty"`
Message string `protobuf:"bytes,2,opt,name=message,proto3" json:"message,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *UpdateWeightsResponse) Reset() {
*x = UpdateWeightsResponse{}
mi := &file_sglang_scheduler_proto_msgTypes[28]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *UpdateWeightsResponse) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*UpdateWeightsResponse) ProtoMessage() {}
func (x *UpdateWeightsResponse) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[28]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use UpdateWeightsResponse.ProtoReflect.Descriptor instead.
func (*UpdateWeightsResponse) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{28}
}
func (x *UpdateWeightsResponse) GetSuccess() bool {
if x != nil {
return x.Success
}
return false
}
func (x *UpdateWeightsResponse) GetMessage() string {
if x != nil {
return x.Message
}
return ""
}
// Get internal state for debugging
type GetInternalStateRequest struct {
state protoimpl.MessageState `protogen:"open.v1"`
StateKeys []string `protobuf:"bytes,1,rep,name=state_keys,json=stateKeys,proto3" json:"state_keys,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *GetInternalStateRequest) Reset() {
*x = GetInternalStateRequest{}
mi := &file_sglang_scheduler_proto_msgTypes[29]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *GetInternalStateRequest) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*GetInternalStateRequest) ProtoMessage() {}
func (x *GetInternalStateRequest) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[29]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use GetInternalStateRequest.ProtoReflect.Descriptor instead.
func (*GetInternalStateRequest) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{29}
}
func (x *GetInternalStateRequest) GetStateKeys() []string {
if x != nil {
return x.StateKeys
}
return nil
}
type GetInternalStateResponse struct {
state protoimpl.MessageState `protogen:"open.v1"`
State *structpb.Struct `protobuf:"bytes,1,opt,name=state,proto3" json:"state,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *GetInternalStateResponse) Reset() {
*x = GetInternalStateResponse{}
mi := &file_sglang_scheduler_proto_msgTypes[30]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *GetInternalStateResponse) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*GetInternalStateResponse) ProtoMessage() {}
func (x *GetInternalStateResponse) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[30]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use GetInternalStateResponse.ProtoReflect.Descriptor instead.
func (*GetInternalStateResponse) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{30}
}
func (x *GetInternalStateResponse) GetState() *structpb.Struct {
if x != nil {
return x.State
}
return nil
}
// Set internal state for testing
type SetInternalStateRequest struct {
state protoimpl.MessageState `protogen:"open.v1"`
State *structpb.Struct `protobuf:"bytes,1,opt,name=state,proto3" json:"state,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *SetInternalStateRequest) Reset() {
*x = SetInternalStateRequest{}
mi := &file_sglang_scheduler_proto_msgTypes[31]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *SetInternalStateRequest) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*SetInternalStateRequest) ProtoMessage() {}
func (x *SetInternalStateRequest) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[31]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use SetInternalStateRequest.ProtoReflect.Descriptor instead.
func (*SetInternalStateRequest) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{31}
}
func (x *SetInternalStateRequest) GetState() *structpb.Struct {
if x != nil {
return x.State
}
return nil
}
type SetInternalStateResponse struct {
state protoimpl.MessageState `protogen:"open.v1"`
Success bool `protobuf:"varint,1,opt,name=success,proto3" json:"success,omitempty"`
Message string `protobuf:"bytes,2,opt,name=message,proto3" json:"message,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *SetInternalStateResponse) Reset() {
*x = SetInternalStateResponse{}
mi := &file_sglang_scheduler_proto_msgTypes[32]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *SetInternalStateResponse) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*SetInternalStateResponse) ProtoMessage() {}
func (x *SetInternalStateResponse) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[32]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use SetInternalStateResponse.ProtoReflect.Descriptor instead.
func (*SetInternalStateResponse) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{32}
}
func (x *SetInternalStateResponse) GetSuccess() bool {
if x != nil {
return x.Success
}
return false
}
func (x *SetInternalStateResponse) GetMessage() string {
if x != nil {
return x.Message
}
return ""
}
// Get model information
type GetModelInfoRequest struct {
state protoimpl.MessageState `protogen:"open.v1"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *GetModelInfoRequest) Reset() {
*x = GetModelInfoRequest{}
mi := &file_sglang_scheduler_proto_msgTypes[33]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *GetModelInfoRequest) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*GetModelInfoRequest) ProtoMessage() {}
func (x *GetModelInfoRequest) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[33]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use GetModelInfoRequest.ProtoReflect.Descriptor instead.
func (*GetModelInfoRequest) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{33}
}
type GetModelInfoResponse struct {
state protoimpl.MessageState `protogen:"open.v1"`
ModelPath string `protobuf:"bytes,1,opt,name=model_path,json=modelPath,proto3" json:"model_path,omitempty"`
TokenizerPath string `protobuf:"bytes,2,opt,name=tokenizer_path,json=tokenizerPath,proto3" json:"tokenizer_path,omitempty"`
IsGeneration bool `protobuf:"varint,3,opt,name=is_generation,json=isGeneration,proto3" json:"is_generation,omitempty"`
PreferredSamplingParams string `protobuf:"bytes,4,opt,name=preferred_sampling_params,json=preferredSamplingParams,proto3" json:"preferred_sampling_params,omitempty"` // JSON string or empty
WeightVersion string `protobuf:"bytes,5,opt,name=weight_version,json=weightVersion,proto3" json:"weight_version,omitempty"`
ServedModelName string `protobuf:"bytes,6,opt,name=served_model_name,json=servedModelName,proto3" json:"served_model_name,omitempty"`
MaxContextLength int32 `protobuf:"varint,7,opt,name=max_context_length,json=maxContextLength,proto3" json:"max_context_length,omitempty"`
VocabSize int32 `protobuf:"varint,8,opt,name=vocab_size,json=vocabSize,proto3" json:"vocab_size,omitempty"`
SupportsVision bool `protobuf:"varint,9,opt,name=supports_vision,json=supportsVision,proto3" json:"supports_vision,omitempty"`
ModelType string `protobuf:"bytes,10,opt,name=model_type,json=modelType,proto3" json:"model_type,omitempty"`
EosTokenIds []int32 `protobuf:"varint,11,rep,packed,name=eos_token_ids,json=eosTokenIds,proto3" json:"eos_token_ids,omitempty"`
PadTokenId int32 `protobuf:"varint,12,opt,name=pad_token_id,json=padTokenId,proto3" json:"pad_token_id,omitempty"`
BosTokenId int32 `protobuf:"varint,13,opt,name=bos_token_id,json=bosTokenId,proto3" json:"bos_token_id,omitempty"`
MaxReqInputLen int32 `protobuf:"varint,14,opt,name=max_req_input_len,json=maxReqInputLen,proto3" json:"max_req_input_len,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *GetModelInfoResponse) Reset() {
*x = GetModelInfoResponse{}
mi := &file_sglang_scheduler_proto_msgTypes[34]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *GetModelInfoResponse) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*GetModelInfoResponse) ProtoMessage() {}
func (x *GetModelInfoResponse) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[34]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use GetModelInfoResponse.ProtoReflect.Descriptor instead.
func (*GetModelInfoResponse) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{34}
}
func (x *GetModelInfoResponse) GetModelPath() string {
if x != nil {
return x.ModelPath
}
return ""
}
func (x *GetModelInfoResponse) GetTokenizerPath() string {
if x != nil {
return x.TokenizerPath
}
return ""
}
func (x *GetModelInfoResponse) GetIsGeneration() bool {
if x != nil {
return x.IsGeneration
}
return false
}
func (x *GetModelInfoResponse) GetPreferredSamplingParams() string {
if x != nil {
return x.PreferredSamplingParams
}
return ""
}
func (x *GetModelInfoResponse) GetWeightVersion() string {
if x != nil {
return x.WeightVersion
}
return ""
}
func (x *GetModelInfoResponse) GetServedModelName() string {
if x != nil {
return x.ServedModelName
}
return ""
}
func (x *GetModelInfoResponse) GetMaxContextLength() int32 {
if x != nil {
return x.MaxContextLength
}
return 0
}
func (x *GetModelInfoResponse) GetVocabSize() int32 {
if x != nil {
return x.VocabSize
}
return 0
}
func (x *GetModelInfoResponse) GetSupportsVision() bool {
if x != nil {
return x.SupportsVision
}
return false
}
func (x *GetModelInfoResponse) GetModelType() string {
if x != nil {
return x.ModelType
}
return ""
}
func (x *GetModelInfoResponse) GetEosTokenIds() []int32 {
if x != nil {
return x.EosTokenIds
}
return nil
}
func (x *GetModelInfoResponse) GetPadTokenId() int32 {
if x != nil {
return x.PadTokenId
}
return 0
}
func (x *GetModelInfoResponse) GetBosTokenId() int32 {
if x != nil {
return x.BosTokenId
}
return 0
}
func (x *GetModelInfoResponse) GetMaxReqInputLen() int32 {
if x != nil {
return x.MaxReqInputLen
}
return 0
}
// Get server information
type GetServerInfoRequest struct {
state protoimpl.MessageState `protogen:"open.v1"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *GetServerInfoRequest) Reset() {
*x = GetServerInfoRequest{}
mi := &file_sglang_scheduler_proto_msgTypes[35]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *GetServerInfoRequest) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*GetServerInfoRequest) ProtoMessage() {}
func (x *GetServerInfoRequest) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[35]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use GetServerInfoRequest.ProtoReflect.Descriptor instead.
func (*GetServerInfoRequest) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{35}
}
type GetServerInfoResponse struct {
state protoimpl.MessageState `protogen:"open.v1"`
// Server configuration (as structured data)
ServerArgs *structpb.Struct `protobuf:"bytes,1,opt,name=server_args,json=serverArgs,proto3" json:"server_args,omitempty"`
// Scheduler metrics (from scheduler initialization)
SchedulerInfo *structpb.Struct `protobuf:"bytes,2,opt,name=scheduler_info,json=schedulerInfo,proto3" json:"scheduler_info,omitempty"`
// Runtime state
ActiveRequests int32 `protobuf:"varint,3,opt,name=active_requests,json=activeRequests,proto3" json:"active_requests,omitempty"`
IsPaused bool `protobuf:"varint,4,opt,name=is_paused,json=isPaused,proto3" json:"is_paused,omitempty"`
LastReceiveTimestamp float64 `protobuf:"fixed64,5,opt,name=last_receive_timestamp,json=lastReceiveTimestamp,proto3" json:"last_receive_timestamp,omitempty"`
UptimeSeconds float64 `protobuf:"fixed64,6,opt,name=uptime_seconds,json=uptimeSeconds,proto3" json:"uptime_seconds,omitempty"`
// Version info
SglangVersion string `protobuf:"bytes,7,opt,name=sglang_version,json=sglangVersion,proto3" json:"sglang_version,omitempty"`
// Server metadata
ServerType string `protobuf:"bytes,8,opt,name=server_type,json=serverType,proto3" json:"server_type,omitempty"` // "grpc"
StartTime *timestamppb.Timestamp `protobuf:"bytes,9,opt,name=start_time,json=startTime,proto3" json:"start_time,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *GetServerInfoResponse) Reset() {
*x = GetServerInfoResponse{}
mi := &file_sglang_scheduler_proto_msgTypes[36]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
func (x *GetServerInfoResponse) String() string {
return protoimpl.X.MessageStringOf(x)
}
func (*GetServerInfoResponse) ProtoMessage() {}
func (x *GetServerInfoResponse) ProtoReflect() protoreflect.Message {
mi := &file_sglang_scheduler_proto_msgTypes[36]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
ms.StoreMessageInfo(mi)
}
return ms
}
return mi.MessageOf(x)
}
// Deprecated: Use GetServerInfoResponse.ProtoReflect.Descriptor instead.
func (*GetServerInfoResponse) Descriptor() ([]byte, []int) {
return file_sglang_scheduler_proto_rawDescGZIP(), []int{36}
}
func (x *GetServerInfoResponse) GetServerArgs() *structpb.Struct {
if x != nil {
return x.ServerArgs
}
return nil
}
func (x *GetServerInfoResponse) GetSchedulerInfo() *structpb.Struct {
if x != nil {
return x.SchedulerInfo
}
return nil
}
func (x *GetServerInfoResponse) GetActiveRequests() int32 {
if x != nil {
return x.ActiveRequests
}
return 0
}
func (x *GetServerInfoResponse) GetIsPaused() bool {
if x != nil {
return x.IsPaused
}
return false
}
func (x *GetServerInfoResponse) GetLastReceiveTimestamp() float64 {
if x != nil {
return x.LastReceiveTimestamp
}
return 0
}
func (x *GetServerInfoResponse) GetUptimeSeconds() float64 {
if x != nil {
return x.UptimeSeconds
}
return 0
}
func (x *GetServerInfoResponse) GetSglangVersion() string {
if x != nil {
return x.SglangVersion
}
return ""
}
func (x *GetServerInfoResponse) GetServerType() string {
if x != nil {
return x.ServerType
}
return ""
}
func (x *GetServerInfoResponse) GetStartTime() *timestamppb.Timestamp {
if x != nil {
return x.StartTime
}
return nil
}
var File_sglang_scheduler_proto protoreflect.FileDescriptor
const file_sglang_scheduler_proto_rawDesc = "" +
"\n" +
"\x16sglang_scheduler.proto\x12\x15sglang.grpc.scheduler\x1a\x1fgoogle/protobuf/timestamp.proto\x1a\x1cgoogle/protobuf/struct.proto\"\x82\b\n" +
"\x0eSamplingParams\x12 \n" +
"\vtemperature\x18\x01 \x01(\x02R\vtemperature\x12\x13\n" +
"\x05top_p\x18\x02 \x01(\x02R\x04topP\x12\x13\n" +
"\x05top_k\x18\x03 \x01(\x05R\x04topK\x12\x13\n" +
"\x05min_p\x18\x04 \x01(\x02R\x04minP\x12+\n" +
"\x11frequency_penalty\x18\x05 \x01(\x02R\x10frequencyPenalty\x12)\n" +
"\x10presence_penalty\x18\x06 \x01(\x02R\x0fpresencePenalty\x12-\n" +
"\x12repetition_penalty\x18\a \x01(\x02R\x11repetitionPenalty\x12)\n" +
"\x0emax_new_tokens\x18\b \x01(\x05H\x01R\fmaxNewTokens\x88\x01\x01\x12\x12\n" +
"\x04stop\x18\t \x03(\tR\x04stop\x12$\n" +
"\x0estop_token_ids\x18\n" +
" \x03(\rR\fstopTokenIds\x12.\n" +
"\x13skip_special_tokens\x18\v \x01(\bR\x11skipSpecialTokens\x12A\n" +
"\x1dspaces_between_special_tokens\x18\f \x01(\bR\x1aspacesBetweenSpecialTokens\x12\x16\n" +
"\x05regex\x18\r \x01(\tH\x00R\x05regex\x12!\n" +
"\vjson_schema\x18\x0e \x01(\tH\x00R\n" +
"jsonSchema\x12#\n" +
"\febnf_grammar\x18\x0f \x01(\tH\x00R\vebnfGrammar\x12'\n" +
"\x0estructural_tag\x18\x10 \x01(\tH\x00R\rstructuralTag\x12\f\n" +
"\x01n\x18\x11 \x01(\x05R\x01n\x12$\n" +
"\x0emin_new_tokens\x18\x12 \x01(\x05R\fminNewTokens\x12\x1d\n" +
"\n" +
"ignore_eos\x18\x13 \x01(\bR\tignoreEos\x12 \n" +
"\fno_stop_trim\x18\x14 \x01(\bR\n" +
"noStopTrim\x12,\n" +
"\x0fstream_interval\x18\x15 \x01(\x05H\x02R\x0estreamInterval\x88\x01\x01\x12S\n" +
"\n" +
"logit_bias\x18\x16 \x03(\v24.sglang.grpc.scheduler.SamplingParams.LogitBiasEntryR\tlogitBias\x12<\n" +
"\rcustom_params\x18\x17 \x01(\v2\x17.google.protobuf.StructR\fcustomParams\x1a<\n" +
"\x0eLogitBiasEntry\x12\x10\n" +
"\x03key\x18\x01 \x01(\tR\x03key\x12\x14\n" +
"\x05value\x18\x02 \x01(\x02R\x05value:\x028\x01B\f\n" +
"\n" +
"constraintB\x11\n" +
"\x0f_max_new_tokensB\x12\n" +
"\x10_stream_interval\"\x8a\x01\n" +
"\x13DisaggregatedParams\x12%\n" +
"\x0ebootstrap_host\x18\x01 \x01(\tR\rbootstrapHost\x12%\n" +
"\x0ebootstrap_port\x18\x02 \x01(\x05R\rbootstrapPort\x12%\n" +
"\x0ebootstrap_room\x18\x03 \x01(\x05R\rbootstrapRoom\"\xd8\x06\n" +
"\x0fGenerateRequest\x12\x1d\n" +
"\n" +
"request_id\x18\x01 \x01(\tR\trequestId\x12C\n" +
"\ttokenized\x18\x02 \x01(\v2%.sglang.grpc.scheduler.TokenizedInputR\ttokenized\x12D\n" +
"\tmm_inputs\x18\x03 \x01(\v2'.sglang.grpc.scheduler.MultimodalInputsR\bmmInputs\x12N\n" +
"\x0fsampling_params\x18\x04 \x01(\v2%.sglang.grpc.scheduler.SamplingParamsR\x0esamplingParams\x12%\n" +
"\x0ereturn_logprob\x18\x05 \x01(\bR\rreturnLogprob\x12*\n" +
"\x11logprob_start_len\x18\x06 \x01(\x05R\x0flogprobStartLen\x12(\n" +
"\x10top_logprobs_num\x18\a \x01(\x05R\x0etopLogprobsNum\x12*\n" +
"\x11token_ids_logprob\x18\b \x03(\rR\x0ftokenIdsLogprob\x120\n" +
"\x14return_hidden_states\x18\t \x01(\bR\x12returnHiddenStates\x12]\n" +
"\x14disaggregated_params\x18\n" +
" \x01(\v2*.sglang.grpc.scheduler.DisaggregatedParamsR\x13disaggregatedParams\x124\n" +
"\x16custom_logit_processor\x18\v \x01(\tR\x14customLogitProcessor\x128\n" +
"\ttimestamp\x18\f \x01(\v2\x1a.google.protobuf.TimestampR\ttimestamp\x12\x1f\n" +
"\vlog_metrics\x18\r \x01(\bR\n" +
"logMetrics\x12!\n" +
"\finput_embeds\x18\x0e \x03(\x02R\vinputEmbeds\x12\x17\n" +
"\alora_id\x18\x0f \x01(\tR\x06loraId\x12,\n" +
"\x12data_parallel_rank\x18\x10 \x01(\x05R\x10dataParallelRank\x12\x16\n" +
"\x06stream\x18\x11 \x01(\bR\x06stream\"R\n" +
"\x0eTokenizedInput\x12#\n" +
"\roriginal_text\x18\x01 \x01(\tR\foriginalText\x12\x1b\n" +
"\tinput_ids\x18\x02 \x03(\rR\binputIds\"\xb4\x02\n" +
"\x10MultimodalInputs\x12\x1d\n" +
"\n" +
"image_urls\x18\x01 \x03(\tR\timageUrls\x12\x1d\n" +
"\n" +
"video_urls\x18\x02 \x03(\tR\tvideoUrls\x12\x1d\n" +
"\n" +
"audio_urls\x18\x03 \x03(\tR\taudioUrls\x12F\n" +
"\x12processed_features\x18\x04 \x01(\v2\x17.google.protobuf.StructR\x11processedFeatures\x12\x1d\n" +
"\n" +
"image_data\x18\x05 \x03(\fR\timageData\x12\x1d\n" +
"\n" +
"video_data\x18\x06 \x03(\fR\tvideoData\x12\x1d\n" +
"\n" +
"audio_data\x18\a \x03(\fR\taudioData\x12\x1e\n" +
"\n" +
"modalities\x18\b \x03(\tR\n" +
"modalities\"\x86\x02\n" +
"\x10GenerateResponse\x12\x1d\n" +
"\n" +
"request_id\x18\x01 \x01(\tR\trequestId\x12B\n" +
"\x05chunk\x18\x02 \x01(\v2*.sglang.grpc.scheduler.GenerateStreamChunkH\x00R\x05chunk\x12E\n" +
"\bcomplete\x18\x03 \x01(\v2'.sglang.grpc.scheduler.GenerateCompleteH\x00R\bcomplete\x12<\n" +
"\x05error\x18\x04 \x01(\v2$.sglang.grpc.scheduler.GenerateErrorH\x00R\x05errorB\n" +
"\n" +
"\bresponse\"\x81\x03\n" +
"\x13GenerateStreamChunk\x12\x1b\n" +
"\ttoken_ids\x18\x01 \x03(\rR\btokenIds\x12#\n" +
"\rprompt_tokens\x18\x02 \x01(\x05R\fpromptTokens\x12+\n" +
"\x11completion_tokens\x18\x03 \x01(\x05R\x10completionTokens\x12#\n" +
"\rcached_tokens\x18\x04 \x01(\x05R\fcachedTokens\x12N\n" +
"\x0foutput_logprobs\x18\x05 \x01(\v2%.sglang.grpc.scheduler.OutputLogProbsR\x0eoutputLogprobs\x12#\n" +
"\rhidden_states\x18\x06 \x03(\x02R\fhiddenStates\x12K\n" +
"\x0einput_logprobs\x18\a \x01(\v2$.sglang.grpc.scheduler.InputLogProbsR\rinputLogprobs\x12\x14\n" +
"\x05index\x18\b \x01(\rR\x05index\"\xb9\x04\n" +
"\x10GenerateComplete\x12\x1d\n" +
"\n" +
"output_ids\x18\x01 \x03(\rR\toutputIds\x12#\n" +
"\rfinish_reason\x18\x02 \x01(\tR\ffinishReason\x12#\n" +
"\rprompt_tokens\x18\x03 \x01(\x05R\fpromptTokens\x12+\n" +
"\x11completion_tokens\x18\x04 \x01(\x05R\x10completionTokens\x12#\n" +
"\rcached_tokens\x18\x05 \x01(\x05R\fcachedTokens\x12N\n" +
"\x0foutput_logprobs\x18\x06 \x01(\v2%.sglang.grpc.scheduler.OutputLogProbsR\x0eoutputLogprobs\x12O\n" +
"\x11all_hidden_states\x18\a \x03(\v2#.sglang.grpc.scheduler.HiddenStatesR\x0fallHiddenStates\x12*\n" +
"\x10matched_token_id\x18\b \x01(\rH\x00R\x0ematchedTokenId\x12*\n" +
"\x10matched_stop_str\x18\t \x01(\tH\x00R\x0ematchedStopStr\x12K\n" +
"\x0einput_logprobs\x18\n" +
" \x01(\v2$.sglang.grpc.scheduler.InputLogProbsR\rinputLogprobs\x12\x14\n" +
"\x05index\x18\v \x01(\rR\x05indexB\x0e\n" +
"\fmatched_stop\"m\n" +
"\rGenerateError\x12\x18\n" +
"\amessage\x18\x01 \x01(\tR\amessage\x12(\n" +
"\x10http_status_code\x18\x02 \x01(\tR\x0ehttpStatusCode\x12\x18\n" +
"\adetails\x18\x03 \x01(\tR\adetails\"\x9b\x01\n" +
"\x0eOutputLogProbs\x12%\n" +
"\x0etoken_logprobs\x18\x01 \x03(\x02R\rtokenLogprobs\x12\x1b\n" +
"\ttoken_ids\x18\x02 \x03(\x05R\btokenIds\x12E\n" +
"\ftop_logprobs\x18\x03 \x03(\v2\".sglang.grpc.scheduler.TopLogProbsR\vtopLogprobs\"\xc4\x01\n" +
"\rInputLogProbs\x12O\n" +
"\x0etoken_logprobs\x18\x01 \x03(\v2(.sglang.grpc.scheduler.InputTokenLogProbR\rtokenLogprobs\x12\x1b\n" +
"\ttoken_ids\x18\x02 \x03(\x05R\btokenIds\x12E\n" +
"\ftop_logprobs\x18\x03 \x03(\v2\".sglang.grpc.scheduler.TopLogProbsR\vtopLogprobs\"8\n" +
"\x11InputTokenLogProb\x12\x19\n" +
"\x05value\x18\x01 \x01(\x02H\x00R\x05value\x88\x01\x01B\b\n" +
"\x06_value\"B\n" +
"\vTopLogProbs\x12\x16\n" +
"\x06values\x18\x01 \x03(\x02R\x06values\x12\x1b\n" +
"\ttoken_ids\x18\x02 \x03(\x05R\btokenIds\"X\n" +
"\fHiddenStates\x12\x16\n" +
"\x06values\x18\x01 \x03(\x02R\x06values\x12\x14\n" +
"\x05layer\x18\x02 \x01(\x05R\x05layer\x12\x1a\n" +
"\bposition\x18\x03 \x01(\x05R\bposition\"\xbd\x03\n" +
"\fEmbedRequest\x12\x1d\n" +
"\n" +
"request_id\x18\x01 \x01(\tR\trequestId\x12C\n" +
"\ttokenized\x18\x02 \x01(\v2%.sglang.grpc.scheduler.TokenizedInputR\ttokenized\x12D\n" +
"\tmm_inputs\x18\x04 \x01(\v2'.sglang.grpc.scheduler.MultimodalInputsR\bmmInputs\x12N\n" +
"\x0fsampling_params\x18\x05 \x01(\v2%.sglang.grpc.scheduler.SamplingParamsR\x0esamplingParams\x12\x1f\n" +
"\vlog_metrics\x18\x06 \x01(\bR\n" +
"logMetrics\x12$\n" +
"\x0etoken_type_ids\x18\a \x03(\x05R\ftokenTypeIds\x12,\n" +
"\x12data_parallel_rank\x18\b \x01(\x05R\x10dataParallelRank\x12(\n" +
"\x10is_cross_encoder\x18\t \x01(\bR\x0eisCrossEncoder\x12\x14\n" +
"\x05texts\x18\n" +
" \x03(\tR\x05texts\"\xb9\x01\n" +
"\rEmbedResponse\x12\x1d\n" +
"\n" +
"request_id\x18\x01 \x01(\tR\trequestId\x12B\n" +
"\bcomplete\x18\x02 \x01(\v2$.sglang.grpc.scheduler.EmbedCompleteH\x00R\bcomplete\x129\n" +
"\x05error\x18\x03 \x01(\v2!.sglang.grpc.scheduler.EmbedErrorH\x00R\x05errorB\n" +
"\n" +
"\bresponse\"\xe9\x01\n" +
"\rEmbedComplete\x12\x1c\n" +
"\tembedding\x18\x01 \x03(\x02R\tembedding\x12#\n" +
"\rprompt_tokens\x18\x02 \x01(\x05R\fpromptTokens\x12#\n" +
"\rcached_tokens\x18\x03 \x01(\x05R\fcachedTokens\x12#\n" +
"\rembedding_dim\x18\x04 \x01(\x05R\fembeddingDim\x12K\n" +
"\x10batch_embeddings\x18\x05 \x03(\v2 .sglang.grpc.scheduler.EmbeddingR\x0fbatchEmbeddings\"9\n" +
"\tEmbedding\x12\x16\n" +
"\x06values\x18\x01 \x03(\x02R\x06values\x12\x14\n" +
"\x05index\x18\x02 \x01(\x05R\x05index\"T\n" +
"\n" +
"EmbedError\x12\x18\n" +
"\amessage\x18\x01 \x01(\tR\amessage\x12\x12\n" +
"\x04code\x18\x02 \x01(\tR\x04code\x12\x18\n" +
"\adetails\x18\x03 \x01(\tR\adetails\"\x14\n" +
"\x12HealthCheckRequest\"I\n" +
"\x13HealthCheckResponse\x12\x18\n" +
"\ahealthy\x18\x01 \x01(\bR\ahealthy\x12\x18\n" +
"\amessage\x18\x02 \x01(\tR\amessage\"E\n" +
"\fAbortRequest\x12\x1d\n" +
"\n" +
"request_id\x18\x01 \x01(\tR\trequestId\x12\x16\n" +
"\x06reason\x18\x02 \x01(\tR\x06reason\"C\n" +
"\rAbortResponse\x12\x18\n" +
"\asuccess\x18\x01 \x01(\bR\asuccess\x12\x18\n" +
"\amessage\x18\x02 \x01(\tR\amessage\"g\n" +
"\x0fLoadLoRARequest\x12\x1d\n" +
"\n" +
"adapter_id\x18\x01 \x01(\tR\tadapterId\x12!\n" +
"\fadapter_path\x18\x02 \x01(\tR\vadapterPath\x12\x12\n" +
"\x04rank\x18\x03 \x01(\x05R\x04rank\"e\n" +
"\x10LoadLoRAResponse\x12\x18\n" +
"\asuccess\x18\x01 \x01(\bR\asuccess\x12\x1d\n" +
"\n" +
"adapter_id\x18\x02 \x01(\tR\tadapterId\x12\x18\n" +
"\amessage\x18\x03 \x01(\tR\amessage\"2\n" +
"\x11UnloadLoRARequest\x12\x1d\n" +
"\n" +
"adapter_id\x18\x01 \x01(\tR\tadapterId\"H\n" +
"\x12UnloadLoRAResponse\x12\x18\n" +
"\asuccess\x18\x01 \x01(\bR\asuccess\x12\x18\n" +
"\amessage\x18\x02 \x01(\tR\amessage\"\xa4\x01\n" +
"\x14UpdateWeightsRequest\x12\x1d\n" +
"\tdisk_path\x18\x01 \x01(\tH\x00R\bdiskPath\x12!\n" +
"\vtensor_data\x18\x02 \x01(\fH\x00R\n" +
"tensorData\x12\x1f\n" +
"\n" +
"remote_url\x18\x03 \x01(\tH\x00R\tremoteUrl\x12\x1f\n" +
"\vweight_name\x18\x04 \x01(\tR\n" +
"weightNameB\b\n" +
"\x06source\"K\n" +
"\x15UpdateWeightsResponse\x12\x18\n" +
"\asuccess\x18\x01 \x01(\bR\asuccess\x12\x18\n" +
"\amessage\x18\x02 \x01(\tR\amessage\"8\n" +
"\x17GetInternalStateRequest\x12\x1d\n" +
"\n" +
"state_keys\x18\x01 \x03(\tR\tstateKeys\"I\n" +
"\x18GetInternalStateResponse\x12-\n" +
"\x05state\x18\x01 \x01(\v2\x17.google.protobuf.StructR\x05state\"H\n" +
"\x17SetInternalStateRequest\x12-\n" +
"\x05state\x18\x01 \x01(\v2\x17.google.protobuf.StructR\x05state\"N\n" +
"\x18SetInternalStateResponse\x12\x18\n" +
"\asuccess\x18\x01 \x01(\bR\asuccess\x12\x18\n" +
"\amessage\x18\x02 \x01(\tR\amessage\"\x15\n" +
"\x13GetModelInfoRequest\"\xb8\x04\n" +
"\x14GetModelInfoResponse\x12\x1d\n" +
"\n" +
"model_path\x18\x01 \x01(\tR\tmodelPath\x12%\n" +
"\x0etokenizer_path\x18\x02 \x01(\tR\rtokenizerPath\x12#\n" +
"\ris_generation\x18\x03 \x01(\bR\fisGeneration\x12:\n" +
"\x19preferred_sampling_params\x18\x04 \x01(\tR\x17preferredSamplingParams\x12%\n" +
"\x0eweight_version\x18\x05 \x01(\tR\rweightVersion\x12*\n" +
"\x11served_model_name\x18\x06 \x01(\tR\x0fservedModelName\x12,\n" +
"\x12max_context_length\x18\a \x01(\x05R\x10maxContextLength\x12\x1d\n" +
"\n" +
"vocab_size\x18\b \x01(\x05R\tvocabSize\x12'\n" +
"\x0fsupports_vision\x18\t \x01(\bR\x0esupportsVision\x12\x1d\n" +
"\n" +
"model_type\x18\n" +
" \x01(\tR\tmodelType\x12\"\n" +
"\reos_token_ids\x18\v \x03(\x05R\veosTokenIds\x12 \n" +
"\fpad_token_id\x18\f \x01(\x05R\n" +
"padTokenId\x12 \n" +
"\fbos_token_id\x18\r \x01(\x05R\n" +
"bosTokenId\x12)\n" +
"\x11max_req_input_len\x18\x0e \x01(\x05R\x0emaxReqInputLen\"\x16\n" +
"\x14GetServerInfoRequest\"\xb7\x03\n" +
"\x15GetServerInfoResponse\x128\n" +
"\vserver_args\x18\x01 \x01(\v2\x17.google.protobuf.StructR\n" +
"serverArgs\x12>\n" +
"\x0escheduler_info\x18\x02 \x01(\v2\x17.google.protobuf.StructR\rschedulerInfo\x12'\n" +
"\x0factive_requests\x18\x03 \x01(\x05R\x0eactiveRequests\x12\x1b\n" +
"\tis_paused\x18\x04 \x01(\bR\bisPaused\x124\n" +
"\x16last_receive_timestamp\x18\x05 \x01(\x01R\x14lastReceiveTimestamp\x12%\n" +
"\x0euptime_seconds\x18\x06 \x01(\x01R\ruptimeSeconds\x12%\n" +
"\x0esglang_version\x18\a \x01(\tR\rsglangVersion\x12\x1f\n" +
"\vserver_type\x18\b \x01(\tR\n" +
"serverType\x129\n" +
"\n" +
"start_time\x18\t \x01(\v2\x1a.google.protobuf.TimestampR\tstartTime2\xd3\x04\n" +
"\x0fSglangScheduler\x12]\n" +
"\bGenerate\x12&.sglang.grpc.scheduler.GenerateRequest\x1a'.sglang.grpc.scheduler.GenerateResponse0\x01\x12R\n" +
"\x05Embed\x12#.sglang.grpc.scheduler.EmbedRequest\x1a$.sglang.grpc.scheduler.EmbedResponse\x12d\n" +
"\vHealthCheck\x12).sglang.grpc.scheduler.HealthCheckRequest\x1a*.sglang.grpc.scheduler.HealthCheckResponse\x12R\n" +
"\x05Abort\x12#.sglang.grpc.scheduler.AbortRequest\x1a$.sglang.grpc.scheduler.AbortResponse\x12g\n" +
"\fGetModelInfo\x12*.sglang.grpc.scheduler.GetModelInfoRequest\x1a+.sglang.grpc.scheduler.GetModelInfoResponse\x12j\n" +
"\rGetServerInfo\x12+.sglang.grpc.scheduler.GetServerInfoRequest\x1a,.sglang.grpc.scheduler.GetServerInfoResponseb\x06proto3"
var (
file_sglang_scheduler_proto_rawDescOnce sync.Once
file_sglang_scheduler_proto_rawDescData []byte
)
func file_sglang_scheduler_proto_rawDescGZIP() []byte {
file_sglang_scheduler_proto_rawDescOnce.Do(func() {
file_sglang_scheduler_proto_rawDescData = protoimpl.X.CompressGZIP(unsafe.Slice(unsafe.StringData(file_sglang_scheduler_proto_rawDesc), len(file_sglang_scheduler_proto_rawDesc)))
})
return file_sglang_scheduler_proto_rawDescData
}
var file_sglang_scheduler_proto_msgTypes = make([]protoimpl.MessageInfo, 38)
var file_sglang_scheduler_proto_goTypes = []any{
(*SamplingParams)(nil), // 0: sglang.grpc.scheduler.SamplingParams
(*DisaggregatedParams)(nil), // 1: sglang.grpc.scheduler.DisaggregatedParams
(*GenerateRequest)(nil), // 2: sglang.grpc.scheduler.GenerateRequest
(*TokenizedInput)(nil), // 3: sglang.grpc.scheduler.TokenizedInput
(*MultimodalInputs)(nil), // 4: sglang.grpc.scheduler.MultimodalInputs
(*GenerateResponse)(nil), // 5: sglang.grpc.scheduler.GenerateResponse
(*GenerateStreamChunk)(nil), // 6: sglang.grpc.scheduler.GenerateStreamChunk
(*GenerateComplete)(nil), // 7: sglang.grpc.scheduler.GenerateComplete
(*GenerateError)(nil), // 8: sglang.grpc.scheduler.GenerateError
(*OutputLogProbs)(nil), // 9: sglang.grpc.scheduler.OutputLogProbs
(*InputLogProbs)(nil), // 10: sglang.grpc.scheduler.InputLogProbs
(*InputTokenLogProb)(nil), // 11: sglang.grpc.scheduler.InputTokenLogProb
(*TopLogProbs)(nil), // 12: sglang.grpc.scheduler.TopLogProbs
(*HiddenStates)(nil), // 13: sglang.grpc.scheduler.HiddenStates
(*EmbedRequest)(nil), // 14: sglang.grpc.scheduler.EmbedRequest
(*EmbedResponse)(nil), // 15: sglang.grpc.scheduler.EmbedResponse
(*EmbedComplete)(nil), // 16: sglang.grpc.scheduler.EmbedComplete
(*Embedding)(nil), // 17: sglang.grpc.scheduler.Embedding
(*EmbedError)(nil), // 18: sglang.grpc.scheduler.EmbedError
(*HealthCheckRequest)(nil), // 19: sglang.grpc.scheduler.HealthCheckRequest
(*HealthCheckResponse)(nil), // 20: sglang.grpc.scheduler.HealthCheckResponse
(*AbortRequest)(nil), // 21: sglang.grpc.scheduler.AbortRequest
(*AbortResponse)(nil), // 22: sglang.grpc.scheduler.AbortResponse
(*LoadLoRARequest)(nil), // 23: sglang.grpc.scheduler.LoadLoRARequest
(*LoadLoRAResponse)(nil), // 24: sglang.grpc.scheduler.LoadLoRAResponse
(*UnloadLoRARequest)(nil), // 25: sglang.grpc.scheduler.UnloadLoRARequest
(*UnloadLoRAResponse)(nil), // 26: sglang.grpc.scheduler.UnloadLoRAResponse
(*UpdateWeightsRequest)(nil), // 27: sglang.grpc.scheduler.UpdateWeightsRequest
(*UpdateWeightsResponse)(nil), // 28: sglang.grpc.scheduler.UpdateWeightsResponse
(*GetInternalStateRequest)(nil), // 29: sglang.grpc.scheduler.GetInternalStateRequest
(*GetInternalStateResponse)(nil), // 30: sglang.grpc.scheduler.GetInternalStateResponse
(*SetInternalStateRequest)(nil), // 31: sglang.grpc.scheduler.SetInternalStateRequest
(*SetInternalStateResponse)(nil), // 32: sglang.grpc.scheduler.SetInternalStateResponse
(*GetModelInfoRequest)(nil), // 33: sglang.grpc.scheduler.GetModelInfoRequest
(*GetModelInfoResponse)(nil), // 34: sglang.grpc.scheduler.GetModelInfoResponse
(*GetServerInfoRequest)(nil), // 35: sglang.grpc.scheduler.GetServerInfoRequest
(*GetServerInfoResponse)(nil), // 36: sglang.grpc.scheduler.GetServerInfoResponse
nil, // 37: sglang.grpc.scheduler.SamplingParams.LogitBiasEntry
(*structpb.Struct)(nil), // 38: google.protobuf.Struct
(*timestamppb.Timestamp)(nil), // 39: google.protobuf.Timestamp
}
var file_sglang_scheduler_proto_depIdxs = []int32{
37, // 0: sglang.grpc.scheduler.SamplingParams.logit_bias:type_name -> sglang.grpc.scheduler.SamplingParams.LogitBiasEntry
38, // 1: sglang.grpc.scheduler.SamplingParams.custom_params:type_name -> google.protobuf.Struct
3, // 2: sglang.grpc.scheduler.GenerateRequest.tokenized:type_name -> sglang.grpc.scheduler.TokenizedInput
4, // 3: sglang.grpc.scheduler.GenerateRequest.mm_inputs:type_name -> sglang.grpc.scheduler.MultimodalInputs
0, // 4: sglang.grpc.scheduler.GenerateRequest.sampling_params:type_name -> sglang.grpc.scheduler.SamplingParams
1, // 5: sglang.grpc.scheduler.GenerateRequest.disaggregated_params:type_name -> sglang.grpc.scheduler.DisaggregatedParams
39, // 6: sglang.grpc.scheduler.GenerateRequest.timestamp:type_name -> google.protobuf.Timestamp
38, // 7: sglang.grpc.scheduler.MultimodalInputs.processed_features:type_name -> google.protobuf.Struct
6, // 8: sglang.grpc.scheduler.GenerateResponse.chunk:type_name -> sglang.grpc.scheduler.GenerateStreamChunk
7, // 9: sglang.grpc.scheduler.GenerateResponse.complete:type_name -> sglang.grpc.scheduler.GenerateComplete
8, // 10: sglang.grpc.scheduler.GenerateResponse.error:type_name -> sglang.grpc.scheduler.GenerateError
9, // 11: sglang.grpc.scheduler.GenerateStreamChunk.output_logprobs:type_name -> sglang.grpc.scheduler.OutputLogProbs
10, // 12: sglang.grpc.scheduler.GenerateStreamChunk.input_logprobs:type_name -> sglang.grpc.scheduler.InputLogProbs
9, // 13: sglang.grpc.scheduler.GenerateComplete.output_logprobs:type_name -> sglang.grpc.scheduler.OutputLogProbs
13, // 14: sglang.grpc.scheduler.GenerateComplete.all_hidden_states:type_name -> sglang.grpc.scheduler.HiddenStates
10, // 15: sglang.grpc.scheduler.GenerateComplete.input_logprobs:type_name -> sglang.grpc.scheduler.InputLogProbs
12, // 16: sglang.grpc.scheduler.OutputLogProbs.top_logprobs:type_name -> sglang.grpc.scheduler.TopLogProbs
11, // 17: sglang.grpc.scheduler.InputLogProbs.token_logprobs:type_name -> sglang.grpc.scheduler.InputTokenLogProb
12, // 18: sglang.grpc.scheduler.InputLogProbs.top_logprobs:type_name -> sglang.grpc.scheduler.TopLogProbs
3, // 19: sglang.grpc.scheduler.EmbedRequest.tokenized:type_name -> sglang.grpc.scheduler.TokenizedInput
4, // 20: sglang.grpc.scheduler.EmbedRequest.mm_inputs:type_name -> sglang.grpc.scheduler.MultimodalInputs
0, // 21: sglang.grpc.scheduler.EmbedRequest.sampling_params:type_name -> sglang.grpc.scheduler.SamplingParams
16, // 22: sglang.grpc.scheduler.EmbedResponse.complete:type_name -> sglang.grpc.scheduler.EmbedComplete
18, // 23: sglang.grpc.scheduler.EmbedResponse.error:type_name -> sglang.grpc.scheduler.EmbedError
17, // 24: sglang.grpc.scheduler.EmbedComplete.batch_embeddings:type_name -> sglang.grpc.scheduler.Embedding
38, // 25: sglang.grpc.scheduler.GetInternalStateResponse.state:type_name -> google.protobuf.Struct
38, // 26: sglang.grpc.scheduler.SetInternalStateRequest.state:type_name -> google.protobuf.Struct
38, // 27: sglang.grpc.scheduler.GetServerInfoResponse.server_args:type_name -> google.protobuf.Struct
38, // 28: sglang.grpc.scheduler.GetServerInfoResponse.scheduler_info:type_name -> google.protobuf.Struct
39, // 29: sglang.grpc.scheduler.GetServerInfoResponse.start_time:type_name -> google.protobuf.Timestamp
2, // 30: sglang.grpc.scheduler.SglangScheduler.Generate:input_type -> sglang.grpc.scheduler.GenerateRequest
14, // 31: sglang.grpc.scheduler.SglangScheduler.Embed:input_type -> sglang.grpc.scheduler.EmbedRequest
19, // 32: sglang.grpc.scheduler.SglangScheduler.HealthCheck:input_type -> sglang.grpc.scheduler.HealthCheckRequest
21, // 33: sglang.grpc.scheduler.SglangScheduler.Abort:input_type -> sglang.grpc.scheduler.AbortRequest
33, // 34: sglang.grpc.scheduler.SglangScheduler.GetModelInfo:input_type -> sglang.grpc.scheduler.GetModelInfoRequest
35, // 35: sglang.grpc.scheduler.SglangScheduler.GetServerInfo:input_type -> sglang.grpc.scheduler.GetServerInfoRequest
5, // 36: sglang.grpc.scheduler.SglangScheduler.Generate:output_type -> sglang.grpc.scheduler.GenerateResponse
15, // 37: sglang.grpc.scheduler.SglangScheduler.Embed:output_type -> sglang.grpc.scheduler.EmbedResponse
20, // 38: sglang.grpc.scheduler.SglangScheduler.HealthCheck:output_type -> sglang.grpc.scheduler.HealthCheckResponse
22, // 39: sglang.grpc.scheduler.SglangScheduler.Abort:output_type -> sglang.grpc.scheduler.AbortResponse
34, // 40: sglang.grpc.scheduler.SglangScheduler.GetModelInfo:output_type -> sglang.grpc.scheduler.GetModelInfoResponse
36, // 41: sglang.grpc.scheduler.SglangScheduler.GetServerInfo:output_type -> sglang.grpc.scheduler.GetServerInfoResponse
36, // [36:42] is the sub-list for method output_type
30, // [30:36] is the sub-list for method input_type
30, // [30:30] is the sub-list for extension type_name
30, // [30:30] is the sub-list for extension extendee
0, // [0:30] is the sub-list for field type_name
}
func init() { file_sglang_scheduler_proto_init() }
func file_sglang_scheduler_proto_init() {
if File_sglang_scheduler_proto != nil {
return
}
file_sglang_scheduler_proto_msgTypes[0].OneofWrappers = []any{
(*SamplingParams_Regex)(nil),
(*SamplingParams_JsonSchema)(nil),
(*SamplingParams_EbnfGrammar)(nil),
(*SamplingParams_StructuralTag)(nil),
}
file_sglang_scheduler_proto_msgTypes[5].OneofWrappers = []any{
(*GenerateResponse_Chunk)(nil),
(*GenerateResponse_Complete)(nil),
(*GenerateResponse_Error)(nil),
}
file_sglang_scheduler_proto_msgTypes[7].OneofWrappers = []any{
(*GenerateComplete_MatchedTokenId)(nil),
(*GenerateComplete_MatchedStopStr)(nil),
}
file_sglang_scheduler_proto_msgTypes[11].OneofWrappers = []any{}
file_sglang_scheduler_proto_msgTypes[15].OneofWrappers = []any{
(*EmbedResponse_Complete)(nil),
(*EmbedResponse_Error)(nil),
}
file_sglang_scheduler_proto_msgTypes[27].OneofWrappers = []any{
(*UpdateWeightsRequest_DiskPath)(nil),
(*UpdateWeightsRequest_TensorData)(nil),
(*UpdateWeightsRequest_RemoteUrl)(nil),
}
type x struct{}
out := protoimpl.TypeBuilder{
File: protoimpl.DescBuilder{
GoPackagePath: reflect.TypeOf(x{}).PkgPath(),
RawDescriptor: unsafe.Slice(unsafe.StringData(file_sglang_scheduler_proto_rawDesc), len(file_sglang_scheduler_proto_rawDesc)),
NumEnums: 0,
NumMessages: 38,
NumExtensions: 0,
NumServices: 1,
},
GoTypes: file_sglang_scheduler_proto_goTypes,
DependencyIndexes: file_sglang_scheduler_proto_depIdxs,
MessageInfos: file_sglang_scheduler_proto_msgTypes,
}.Build()
File_sglang_scheduler_proto = out.File
file_sglang_scheduler_proto_goTypes = nil
file_sglang_scheduler_proto_depIdxs = nil
}
Reference in New Issue View Git Blame Copy Permalink