[router] add get server info and get model info in grpc server (#11303)

python/sglang/srt/entrypoints/grpc_server.py

@@ -5,6 +5,7 @@ Uses GrpcRequestManager for orchestration without tokenization.
 
 import argparse
 import asyncio
+import dataclasses
 import logging
 import multiprocessing as mp
 import os
@@ -15,8 +16,11 @@ from typing import AsyncIterator, Dict, Optional, Tuple
 
 import grpc
 from google.protobuf.json_format import MessageToDict
+from google.protobuf.struct_pb2 import Struct
+from google.protobuf.timestamp_pb2 import Timestamp
 from grpc_reflection.v1alpha import reflection
 
+import sglang
 from sglang.srt.disaggregation.utils import FAKE_BOOTSTRAP_HOST, DisaggregationMode
 from sglang.srt.entrypoints.grpc_request_manager import GrpcRequestManager
 from sglang.srt.grpc import sglang_scheduler_pb2, sglang_scheduler_pb2_grpc
@@ -173,11 +177,13 @@ class SGLangSchedulerServicer(sglang_scheduler_pb2_grpc.SglangSchedulerServicer)
         request_manager: GrpcRequestManager,
         server_args: ServerArgs,
         model_info: Dict,
+        scheduler_info: Dict,
     ):
         """Initialize the standalone gRPC service."""
         self.request_manager = request_manager
         self.server_args = server_args
         self.model_info = model_info
+        self.scheduler_info = scheduler_info
         self.start_time = time.time()
 
         # Start the request manager's event loop using auto_create_handle_loop
@@ -396,6 +402,89 @@ class SGLangSchedulerServicer(sglang_scheduler_pb2_grpc.SglangSchedulerServicer)
                 message=str(e),
             )
 
+    async def GetModelInfo(
+        self,
+        request: sglang_scheduler_pb2.GetModelInfoRequest,
+        context: grpc.aio.ServicerContext,
+    ) -> sglang_scheduler_pb2.GetModelInfoResponse:
+        """Get model information."""
+        logger.info("Model info request received")
+
+        is_generation = self.scheduler_info.get("is_generation")
+        if is_generation is None:
+            is_generation = not self.server_args.is_embedding
+
+        return sglang_scheduler_pb2.GetModelInfoResponse(
+            model_path=self.server_args.model_path,
+            tokenizer_path=self.server_args.tokenizer_path or "",
+            is_generation=is_generation,
+            preferred_sampling_params=(
+                self.server_args.preferred_sampling_params or ""
+            ),
+            weight_version=self.server_args.weight_version or "",
+            served_model_name=self.server_args.served_model_name,
+            max_context_length=self.model_info["max_context_length"],
+            vocab_size=self.model_info["vocab_size"],
+            supports_vision=self.model_info["supports_vision"],
+            model_type=self.model_info["model_type"],
+            eos_token_ids=self.model_info["eos_token_ids"],
+            pad_token_id=self.model_info["pad_token_id"],
+            bos_token_id=self.model_info["bos_token_id"],
+            max_req_input_len=self.model_info["max_req_input_len"],
+        )
+
+    async def GetServerInfo(
+        self,
+        request: sglang_scheduler_pb2.GetServerInfoRequest,
+        context: grpc.aio.ServicerContext,
+    ) -> sglang_scheduler_pb2.GetServerInfoResponse:
+        """Get server information."""
+        logger.info("Server info request received")
+
+        server_args_dict = dataclasses.asdict(self.server_args)
+        server_args_struct = Struct()
+
+        def make_serializable(obj):
+            if obj is None:
+                return None
+            elif isinstance(obj, (str, int, float, bool)):
+                return obj
+            elif isinstance(obj, (list, tuple, set)):
+                return [make_serializable(item) for item in obj]
+            elif isinstance(obj, dict):
+                return {k: make_serializable(v) for k, v in obj.items()}
+            else:
+                return str(obj)
+
+        serializable_args = make_serializable(server_args_dict)
+        server_args_struct.update(serializable_args)
+
+        # Convert scheduler_info to Struct
+        scheduler_info_struct = Struct()
+        scheduler_info_struct.update(self.scheduler_info)
+
+        # Get runtime state from request manager
+        manager_state = self.request_manager.get_server_info()
+
+        # Calculate uptime
+        uptime = time.time() - self.start_time
+
+        # Create timestamp
+        start_timestamp = Timestamp()
+        start_timestamp.FromSeconds(int(self.start_time))
+
+        return sglang_scheduler_pb2.GetServerInfoResponse(
+            server_args=server_args_struct,
+            scheduler_info=scheduler_info_struct,
+            active_requests=manager_state["active_requests"],
+            is_paused=manager_state["paused"],
+            last_receive_timestamp=manager_state["last_receive_time"],
+            uptime_seconds=uptime,
+            sglang_version=sglang.__version__,
+            server_type="grpc",
+            start_time=start_timestamp,
+        )
+
     # Helper methods for request/response conversion
 
     def _convert_generate_request(
@@ -756,6 +845,7 @@ async def serve_grpc(
         request_manager=request_manager,
         server_args=server_args,
         model_info=model_info,
+        scheduler_info=scheduler_info,
     )
     sglang_scheduler_pb2_grpc.add_SglangSchedulerServicer_to_server(servicer, server)
 

python/sglang/srt/grpc/sglang_scheduler.proto

@@ -20,6 +20,12 @@ service SglangScheduler {
   // Abort a running request
   rpc Abort(AbortRequest) returns (AbortResponse);
 
+  // Get model information
+  rpc GetModelInfo(GetModelInfoRequest) returns (GetModelInfoResponse);
+
+  // Get server information
+  rpc GetServerInfo(GetServerInfoRequest) returns (GetServerInfoResponse);
+
 }
 
 // =====================
@@ -401,3 +407,56 @@ message SetInternalStateResponse {
   bool success = 1;
   string message = 2;
 }
+
+// =====================
+// Model and Server Info
+// =====================
+
+// Get model information
+message GetModelInfoRequest {}
+
+message GetModelInfoResponse {
+  string model_path = 1;
+  string tokenizer_path = 2;
+  bool is_generation = 3;
+  string preferred_sampling_params = 4;  // JSON string or empty
+  string weight_version = 5;
+  string served_model_name = 6;
+  int32 max_context_length = 7;
+  int32 vocab_size = 8;
+  bool supports_vision = 9;
+  string model_type = 10;
+  repeated int32 eos_token_ids = 11;
+  int32 pad_token_id = 12;
+  int32 bos_token_id = 13;
+  int32 max_req_input_len = 14;
+}
+
+// Get server information
+message GetServerInfoRequest {}
+
+message GetServerInfoResponse {
+  // Server configuration (as structured data)
+  google.protobuf.Struct server_args = 1;
+
+  // Scheduler metrics (from scheduler initialization)
+  google.protobuf.Struct scheduler_info = 2;
+
+  // Runtime state
+  int32 active_requests = 3;
+  bool is_paused = 4;
+  double last_receive_timestamp = 5;
+  double uptime_seconds = 6;
+
+  // Version info
+  string sglang_version = 7;
+
+  // Server metadata
+  string server_type = 8;  // "grpc"
+  google.protobuf.Timestamp start_time = 9;
+
+  // Note: internal_states not provided in gRPC mode
+  // Scheduler-side metrics (memory usage, throughput) require
+  // bidirectional communicator infrastructure not available in gRPC.
+  // Use HTTP /get_server_info if scheduler internal state is needed.
+}

python/sglang/srt/grpc/sglang_scheduler_pb2.py

@@ -29,7 +29,7 @@ from google.protobuf import timestamp_pb2 as google_dot_protobuf_dot_timestamp__
 from google.protobuf import struct_pb2 as google_dot_protobuf_dot_struct__pb2
 
 
-DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n\x16sglang_scheduler.proto\x12\x15sglang.grpc.scheduler\x1a\x1fgoogle/protobuf/timestamp.proto\x1a\x1cgoogle/protobuf/struct.proto\"\xd0\x05\n\x0eSamplingParams\x12\x13\n\x0btemperature\x18\x01 \x01(\x02\x12\r\n\x05top_p\x18\x02 \x01(\x02\x12\r\n\x05top_k\x18\x03 \x01(\x05\x12\r\n\x05min_p\x18\x04 \x01(\x02\x12\x19\n\x11\x66requency_penalty\x18\x05 \x01(\x02\x12\x18\n\x10presence_penalty\x18\x06 \x01(\x02\x12\x1a\n\x12repetition_penalty\x18\x07 \x01(\x02\x12\x1b\n\x0emax_new_tokens\x18\x08 \x01(\x05H\x01\x88\x01\x01\x12\x0c\n\x04stop\x18\t \x03(\t\x12\x16\n\x0estop_token_ids\x18\n \x03(\r\x12\x1b\n\x13skip_special_tokens\x18\x0b \x01(\x08\x12%\n\x1dspaces_between_special_tokens\x18\x0c \x01(\x08\x12\x0f\n\x05regex\x18\r \x01(\tH\x00\x12\x15\n\x0bjson_schema\x18\x0e \x01(\tH\x00\x12\x16\n\x0c\x65\x62nf_grammar\x18\x0f \x01(\tH\x00\x12\x18\n\x0estructural_tag\x18\x10 \x01(\tH\x00\x12\t\n\x01n\x18\x11 \x01(\x05\x12\x16\n\x0emin_new_tokens\x18\x12 \x01(\x05\x12\x12\n\nignore_eos\x18\x13 \x01(\x08\x12\x14\n\x0cno_stop_trim\x18\x14 \x01(\x08\x12\x1c\n\x0fstream_interval\x18\x15 \x01(\x05H\x02\x88\x01\x01\x12H\n\nlogit_bias\x18\x16 \x03(\x0b\x32\x34.sglang.grpc.scheduler.SamplingParams.LogitBiasEntry\x12.\n\rcustom_params\x18\x17 \x01(\x0b\x32\x17.google.protobuf.Struct\x1a\x30\n\x0eLogitBiasEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\x02:\x02\x38\x01\x42\x0c\n\nconstraintB\x11\n\x0f_max_new_tokensB\x12\n\x10_stream_interval\"]\n\x13\x44isaggregatedParams\x12\x16\n\x0e\x62ootstrap_host\x18\x01 \x01(\t\x12\x16\n\x0e\x62ootstrap_port\x18\x02 \x01(\x05\x12\x16\n\x0e\x62ootstrap_room\x18\x03 \x01(\x05\"\xe2\x04\n\x0fGenerateRequest\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12\x38\n\ttokenized\x18\x02 \x01(\x0b\x32%.sglang.grpc.scheduler.TokenizedInput\x12:\n\tmm_inputs\x18\x03 \x01(\x0b\x32\'.sglang.grpc.scheduler.MultimodalInputs\x12>\n\x0fsampling_params\x18\x04 \x01(\x0b\x32%.sglang.grpc.scheduler.SamplingParams\x12\x16\n\x0ereturn_logprob\x18\x05 \x01(\x08\x12\x19\n\x11logprob_start_len\x18\x06 \x01(\x05\x12\x18\n\x10top_logprobs_num\x18\x07 \x01(\x05\x12\x19\n\x11token_ids_logprob\x18\x08 \x03(\r\x12\x1c\n\x14return_hidden_states\x18\t \x01(\x08\x12H\n\x14\x64isaggregated_params\x18\n \x01(\x0b\x32*.sglang.grpc.scheduler.DisaggregatedParams\x12\x1e\n\x16\x63ustom_logit_processor\x18\x0b \x01(\t\x12-\n\ttimestamp\x18\x0c \x01(\x0b\x32\x1a.google.protobuf.Timestamp\x12\x13\n\x0blog_metrics\x18\r \x01(\x08\x12\x14\n\x0cinput_embeds\x18\x0e \x03(\x02\x12\x0f\n\x07lora_id\x18\x0f \x01(\t\x12\x1a\n\x12\x64\x61ta_parallel_rank\x18\x10 \x01(\x05\x12\x0e\n\x06stream\x18\x11 \x01(\x08\":\n\x0eTokenizedInput\x12\x15\n\roriginal_text\x18\x01 \x01(\t\x12\x11\n\tinput_ids\x18\x02 \x03(\r\"\xd3\x01\n\x10MultimodalInputs\x12\x12\n\nimage_urls\x18\x01 \x03(\t\x12\x12\n\nvideo_urls\x18\x02 \x03(\t\x12\x12\n\naudio_urls\x18\x03 \x03(\t\x12\x33\n\x12processed_features\x18\x04 \x01(\x0b\x32\x17.google.protobuf.Struct\x12\x12\n\nimage_data\x18\x05 \x03(\x0c\x12\x12\n\nvideo_data\x18\x06 \x03(\x0c\x12\x12\n\naudio_data\x18\x07 \x03(\x0c\x12\x12\n\nmodalities\x18\x08 \x03(\t\"\xe3\x01\n\x10GenerateResponse\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12;\n\x05\x63hunk\x18\x02 \x01(\x0b\x32*.sglang.grpc.scheduler.GenerateStreamChunkH\x00\x12;\n\x08\x63omplete\x18\x03 \x01(\x0b\x32\'.sglang.grpc.scheduler.GenerateCompleteH\x00\x12\x35\n\x05\x65rror\x18\x04 \x01(\x0b\x32$.sglang.grpc.scheduler.GenerateErrorH\x00\x42\n\n\x08response\"\x95\x02\n\x13GenerateStreamChunk\x12\x11\n\ttoken_ids\x18\x01 \x03(\r\x12\x15\n\rprompt_tokens\x18\x02 \x01(\x05\x12\x19\n\x11\x63ompletion_tokens\x18\x03 \x01(\x05\x12\x15\n\rcached_tokens\x18\x04 \x01(\x05\x12>\n\x0foutput_logprobs\x18\x05 \x01(\x0b\x32%.sglang.grpc.scheduler.OutputLogProbs\x12\x15\n\rhidden_states\x18\x06 \x03(\x02\x12<\n\x0einput_logprobs\x18\x07 \x01(\x0b\x32$.sglang.grpc.scheduler.InputLogProbs\x12\r\n\x05index\x18\x08 \x01(\r\"\x9b\x03\n\x10GenerateComplete\x12\x12\n\noutput_ids\x18\x01 \x03(\r\x12\x15\n\rfinish_reason\x18\x02 \x01(\t\x12\x15\n\rprompt_tokens\x18\x03 \x01(\x05\x12\x19\n\x11\x63ompletion_tokens\x18\x04 \x01(\x05\x12\x15\n\rcached_tokens\x18\x05 \x01(\x05\x12>\n\x0foutput_logprobs\x18\x06 \x01(\x0b\x32%.sglang.grpc.scheduler.OutputLogProbs\x12>\n\x11\x61ll_hidden_states\x18\x07 \x03(\x0b\x32#.sglang.grpc.scheduler.HiddenStates\x12\x1a\n\x10matched_token_id\x18\x08 \x01(\rH\x00\x12\x1a\n\x10matched_stop_str\x18\t \x01(\tH\x00\x12<\n\x0einput_logprobs\x18\n \x01(\x0b\x32$.sglang.grpc.scheduler.InputLogProbs\x12\r\n\x05index\x18\x0b \x01(\rB\x0e\n\x0cmatched_stop\"K\n\rGenerateError\x12\x0f\n\x07message\x18\x01 \x01(\t\x12\x18\n\x10http_status_code\x18\x02 \x01(\t\x12\x0f\n\x07\x64\x65tails\x18\x03 \x01(\t\"u\n\x0eOutputLogProbs\x12\x16\n\x0etoken_logprobs\x18\x01 \x03(\x02\x12\x11\n\ttoken_ids\x18\x02 \x03(\x05\x12\x38\n\x0ctop_logprobs\x18\x03 \x03(\x0b\x32\".sglang.grpc.scheduler.TopLogProbs\"\x9e\x01\n\rInputLogProbs\x12@\n\x0etoken_logprobs\x18\x01 \x03(\x0b\x32(.sglang.grpc.scheduler.InputTokenLogProb\x12\x11\n\ttoken_ids\x18\x02 \x03(\x05\x12\x38\n\x0ctop_logprobs\x18\x03 \x03(\x0b\x32\".sglang.grpc.scheduler.TopLogProbs\"1\n\x11InputTokenLogProb\x12\x12\n\x05value\x18\x01 \x01(\x02H\x00\x88\x01\x01\x42\x08\n\x06_value\"0\n\x0bTopLogProbs\x12\x0e\n\x06values\x18\x01 \x03(\x02\x12\x11\n\ttoken_ids\x18\x02 \x03(\x05\"?\n\x0cHiddenStates\x12\x0e\n\x06values\x18\x01 \x03(\x02\x12\r\n\x05layer\x18\x02 \x01(\x05\x12\x10\n\x08position\x18\x03 \x01(\x05\"\xca\x02\n\x0c\x45mbedRequest\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12\x38\n\ttokenized\x18\x02 \x01(\x0b\x32%.sglang.grpc.scheduler.TokenizedInput\x12:\n\tmm_inputs\x18\x04 \x01(\x0b\x32\'.sglang.grpc.scheduler.MultimodalInputs\x12>\n\x0fsampling_params\x18\x05 \x01(\x0b\x32%.sglang.grpc.scheduler.SamplingParams\x12\x13\n\x0blog_metrics\x18\x06 \x01(\x08\x12\x16\n\x0etoken_type_ids\x18\x07 \x03(\x05\x12\x1a\n\x12\x64\x61ta_parallel_rank\x18\x08 \x01(\x05\x12\x18\n\x10is_cross_encoder\x18\t \x01(\x08\x12\r\n\x05texts\x18\n \x03(\t\"\x9d\x01\n\rEmbedResponse\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12\x38\n\x08\x63omplete\x18\x02 \x01(\x0b\x32$.sglang.grpc.scheduler.EmbedCompleteH\x00\x12\x32\n\x05\x65rror\x18\x03 \x01(\x0b\x32!.sglang.grpc.scheduler.EmbedErrorH\x00\x42\n\n\x08response\"\xa3\x01\n\rEmbedComplete\x12\x11\n\tembedding\x18\x01 \x03(\x02\x12\x15\n\rprompt_tokens\x18\x02 \x01(\x05\x12\x15\n\rcached_tokens\x18\x03 \x01(\x05\x12\x15\n\rembedding_dim\x18\x04 \x01(\x05\x12:\n\x10\x62\x61tch_embeddings\x18\x05 \x03(\x0b\x32 .sglang.grpc.scheduler.Embedding\"*\n\tEmbedding\x12\x0e\n\x06values\x18\x01 \x03(\x02\x12\r\n\x05index\x18\x02 \x01(\x05\"<\n\nEmbedError\x12\x0f\n\x07message\x18\x01 \x01(\t\x12\x0c\n\x04\x63ode\x18\x02 \x01(\t\x12\x0f\n\x07\x64\x65tails\x18\x03 \x01(\t\"N\n\x12HealthCheckRequest\x12\x38\n\ttokenized\x18\x01 \x01(\x0b\x32%.sglang.grpc.scheduler.TokenizedInput\"7\n\x13HealthCheckResponse\x12\x0f\n\x07healthy\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\"2\n\x0c\x41\x62ortRequest\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12\x0e\n\x06reason\x18\x02 \x01(\t\"1\n\rAbortResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\"I\n\x0fLoadLoRARequest\x12\x12\n\nadapter_id\x18\x01 \x01(\t\x12\x14\n\x0c\x61\x64\x61pter_path\x18\x02 \x01(\t\x12\x0c\n\x04rank\x18\x03 \x01(\x05\"H\n\x10LoadLoRAResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x12\n\nadapter_id\x18\x02 \x01(\t\x12\x0f\n\x07message\x18\x03 \x01(\t\"\'\n\x11UnloadLoRARequest\x12\x12\n\nadapter_id\x18\x01 \x01(\t\"6\n\x12UnloadLoRAResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\"w\n\x14UpdateWeightsRequest\x12\x13\n\tdisk_path\x18\x01 \x01(\tH\x00\x12\x15\n\x0btensor_data\x18\x02 \x01(\x0cH\x00\x12\x14\n\nremote_url\x18\x03 \x01(\tH\x00\x12\x13\n\x0bweight_name\x18\x04 \x01(\tB\x08\n\x06source\"9\n\x15UpdateWeightsResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\"-\n\x17GetInternalStateRequest\x12\x12\n\nstate_keys\x18\x01 \x03(\t\"B\n\x18GetInternalStateResponse\x12&\n\x05state\x18\x01 \x01(\x0b\x32\x17.google.protobuf.Struct\"A\n\x17SetInternalStateRequest\x12&\n\x05state\x18\x01 \x01(\x0b\x32\x17.google.protobuf.Struct\"<\n\x18SetInternalStateResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t2\xfe\x02\n\x0fSglangScheduler\x12]\n\x08Generate\x12&.sglang.grpc.scheduler.GenerateRequest\x1a\'.sglang.grpc.scheduler.GenerateResponse0\x01\x12R\n\x05\x45mbed\x12#.sglang.grpc.scheduler.EmbedRequest\x1a$.sglang.grpc.scheduler.EmbedResponse\x12\x64\n\x0bHealthCheck\x12).sglang.grpc.scheduler.HealthCheckRequest\x1a*.sglang.grpc.scheduler.HealthCheckResponse\x12R\n\x05\x41\x62ort\x12#.sglang.grpc.scheduler.AbortRequest\x1a$.sglang.grpc.scheduler.AbortResponseb\x06proto3')
+DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n\x16sglang_scheduler.proto\x12\x15sglang.grpc.scheduler\x1a\x1fgoogle/protobuf/timestamp.proto\x1a\x1cgoogle/protobuf/struct.proto\"\xd0\x05\n\x0eSamplingParams\x12\x13\n\x0btemperature\x18\x01 \x01(\x02\x12\r\n\x05top_p\x18\x02 \x01(\x02\x12\r\n\x05top_k\x18\x03 \x01(\x05\x12\r\n\x05min_p\x18\x04 \x01(\x02\x12\x19\n\x11\x66requency_penalty\x18\x05 \x01(\x02\x12\x18\n\x10presence_penalty\x18\x06 \x01(\x02\x12\x1a\n\x12repetition_penalty\x18\x07 \x01(\x02\x12\x1b\n\x0emax_new_tokens\x18\x08 \x01(\x05H\x01\x88\x01\x01\x12\x0c\n\x04stop\x18\t \x03(\t\x12\x16\n\x0estop_token_ids\x18\n \x03(\r\x12\x1b\n\x13skip_special_tokens\x18\x0b \x01(\x08\x12%\n\x1dspaces_between_special_tokens\x18\x0c \x01(\x08\x12\x0f\n\x05regex\x18\r \x01(\tH\x00\x12\x15\n\x0bjson_schema\x18\x0e \x01(\tH\x00\x12\x16\n\x0c\x65\x62nf_grammar\x18\x0f \x01(\tH\x00\x12\x18\n\x0estructural_tag\x18\x10 \x01(\tH\x00\x12\t\n\x01n\x18\x11 \x01(\x05\x12\x16\n\x0emin_new_tokens\x18\x12 \x01(\x05\x12\x12\n\nignore_eos\x18\x13 \x01(\x08\x12\x14\n\x0cno_stop_trim\x18\x14 \x01(\x08\x12\x1c\n\x0fstream_interval\x18\x15 \x01(\x05H\x02\x88\x01\x01\x12H\n\nlogit_bias\x18\x16 \x03(\x0b\x32\x34.sglang.grpc.scheduler.SamplingParams.LogitBiasEntry\x12.\n\rcustom_params\x18\x17 \x01(\x0b\x32\x17.google.protobuf.Struct\x1a\x30\n\x0eLogitBiasEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\x02:\x02\x38\x01\x42\x0c\n\nconstraintB\x11\n\x0f_max_new_tokensB\x12\n\x10_stream_interval\"]\n\x13\x44isaggregatedParams\x12\x16\n\x0e\x62ootstrap_host\x18\x01 \x01(\t\x12\x16\n\x0e\x62ootstrap_port\x18\x02 \x01(\x05\x12\x16\n\x0e\x62ootstrap_room\x18\x03 \x01(\x05\"\xe2\x04\n\x0fGenerateRequest\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12\x38\n\ttokenized\x18\x02 \x01(\x0b\x32%.sglang.grpc.scheduler.TokenizedInput\x12:\n\tmm_inputs\x18\x03 \x01(\x0b\x32\'.sglang.grpc.scheduler.MultimodalInputs\x12>\n\x0fsampling_params\x18\x04 \x01(\x0b\x32%.sglang.grpc.scheduler.SamplingParams\x12\x16\n\x0ereturn_logprob\x18\x05 \x01(\x08\x12\x19\n\x11logprob_start_len\x18\x06 \x01(\x05\x12\x18\n\x10top_logprobs_num\x18\x07 \x01(\x05\x12\x19\n\x11token_ids_logprob\x18\x08 \x03(\r\x12\x1c\n\x14return_hidden_states\x18\t \x01(\x08\x12H\n\x14\x64isaggregated_params\x18\n \x01(\x0b\x32*.sglang.grpc.scheduler.DisaggregatedParams\x12\x1e\n\x16\x63ustom_logit_processor\x18\x0b \x01(\t\x12-\n\ttimestamp\x18\x0c \x01(\x0b\x32\x1a.google.protobuf.Timestamp\x12\x13\n\x0blog_metrics\x18\r \x01(\x08\x12\x14\n\x0cinput_embeds\x18\x0e \x03(\x02\x12\x0f\n\x07lora_id\x18\x0f \x01(\t\x12\x1a\n\x12\x64\x61ta_parallel_rank\x18\x10 \x01(\x05\x12\x0e\n\x06stream\x18\x11 \x01(\x08\":\n\x0eTokenizedInput\x12\x15\n\roriginal_text\x18\x01 \x01(\t\x12\x11\n\tinput_ids\x18\x02 \x03(\r\"\xd3\x01\n\x10MultimodalInputs\x12\x12\n\nimage_urls\x18\x01 \x03(\t\x12\x12\n\nvideo_urls\x18\x02 \x03(\t\x12\x12\n\naudio_urls\x18\x03 \x03(\t\x12\x33\n\x12processed_features\x18\x04 \x01(\x0b\x32\x17.google.protobuf.Struct\x12\x12\n\nimage_data\x18\x05 \x03(\x0c\x12\x12\n\nvideo_data\x18\x06 \x03(\x0c\x12\x12\n\naudio_data\x18\x07 \x03(\x0c\x12\x12\n\nmodalities\x18\x08 \x03(\t\"\xe3\x01\n\x10GenerateResponse\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12;\n\x05\x63hunk\x18\x02 \x01(\x0b\x32*.sglang.grpc.scheduler.GenerateStreamChunkH\x00\x12;\n\x08\x63omplete\x18\x03 \x01(\x0b\x32\'.sglang.grpc.scheduler.GenerateCompleteH\x00\x12\x35\n\x05\x65rror\x18\x04 \x01(\x0b\x32$.sglang.grpc.scheduler.GenerateErrorH\x00\x42\n\n\x08response\"\x95\x02\n\x13GenerateStreamChunk\x12\x11\n\ttoken_ids\x18\x01 \x03(\r\x12\x15\n\rprompt_tokens\x18\x02 \x01(\x05\x12\x19\n\x11\x63ompletion_tokens\x18\x03 \x01(\x05\x12\x15\n\rcached_tokens\x18\x04 \x01(\x05\x12>\n\x0foutput_logprobs\x18\x05 \x01(\x0b\x32%.sglang.grpc.scheduler.OutputLogProbs\x12\x15\n\rhidden_states\x18\x06 \x03(\x02\x12<\n\x0einput_logprobs\x18\x07 \x01(\x0b\x32$.sglang.grpc.scheduler.InputLogProbs\x12\r\n\x05index\x18\x08 \x01(\r\"\x9b\x03\n\x10GenerateComplete\x12\x12\n\noutput_ids\x18\x01 \x03(\r\x12\x15\n\rfinish_reason\x18\x02 \x01(\t\x12\x15\n\rprompt_tokens\x18\x03 \x01(\x05\x12\x19\n\x11\x63ompletion_tokens\x18\x04 \x01(\x05\x12\x15\n\rcached_tokens\x18\x05 \x01(\x05\x12>\n\x0foutput_logprobs\x18\x06 \x01(\x0b\x32%.sglang.grpc.scheduler.OutputLogProbs\x12>\n\x11\x61ll_hidden_states\x18\x07 \x03(\x0b\x32#.sglang.grpc.scheduler.HiddenStates\x12\x1a\n\x10matched_token_id\x18\x08 \x01(\rH\x00\x12\x1a\n\x10matched_stop_str\x18\t \x01(\tH\x00\x12<\n\x0einput_logprobs\x18\n \x01(\x0b\x32$.sglang.grpc.scheduler.InputLogProbs\x12\r\n\x05index\x18\x0b \x01(\rB\x0e\n\x0cmatched_stop\"K\n\rGenerateError\x12\x0f\n\x07message\x18\x01 \x01(\t\x12\x18\n\x10http_status_code\x18\x02 \x01(\t\x12\x0f\n\x07\x64\x65tails\x18\x03 \x01(\t\"u\n\x0eOutputLogProbs\x12\x16\n\x0etoken_logprobs\x18\x01 \x03(\x02\x12\x11\n\ttoken_ids\x18\x02 \x03(\x05\x12\x38\n\x0ctop_logprobs\x18\x03 \x03(\x0b\x32\".sglang.grpc.scheduler.TopLogProbs\"\x9e\x01\n\rInputLogProbs\x12@\n\x0etoken_logprobs\x18\x01 \x03(\x0b\x32(.sglang.grpc.scheduler.InputTokenLogProb\x12\x11\n\ttoken_ids\x18\x02 \x03(\x05\x12\x38\n\x0ctop_logprobs\x18\x03 \x03(\x0b\x32\".sglang.grpc.scheduler.TopLogProbs\"1\n\x11InputTokenLogProb\x12\x12\n\x05value\x18\x01 \x01(\x02H\x00\x88\x01\x01\x42\x08\n\x06_value\"0\n\x0bTopLogProbs\x12\x0e\n\x06values\x18\x01 \x03(\x02\x12\x11\n\ttoken_ids\x18\x02 \x03(\x05\"?\n\x0cHiddenStates\x12\x0e\n\x06values\x18\x01 \x03(\x02\x12\r\n\x05layer\x18\x02 \x01(\x05\x12\x10\n\x08position\x18\x03 \x01(\x05\"\xca\x02\n\x0c\x45mbedRequest\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12\x38\n\ttokenized\x18\x02 \x01(\x0b\x32%.sglang.grpc.scheduler.TokenizedInput\x12:\n\tmm_inputs\x18\x04 \x01(\x0b\x32\'.sglang.grpc.scheduler.MultimodalInputs\x12>\n\x0fsampling_params\x18\x05 \x01(\x0b\x32%.sglang.grpc.scheduler.SamplingParams\x12\x13\n\x0blog_metrics\x18\x06 \x01(\x08\x12\x16\n\x0etoken_type_ids\x18\x07 \x03(\x05\x12\x1a\n\x12\x64\x61ta_parallel_rank\x18\x08 \x01(\x05\x12\x18\n\x10is_cross_encoder\x18\t \x01(\x08\x12\r\n\x05texts\x18\n \x03(\t\"\x9d\x01\n\rEmbedResponse\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12\x38\n\x08\x63omplete\x18\x02 \x01(\x0b\x32$.sglang.grpc.scheduler.EmbedCompleteH\x00\x12\x32\n\x05\x65rror\x18\x03 \x01(\x0b\x32!.sglang.grpc.scheduler.EmbedErrorH\x00\x42\n\n\x08response\"\xa3\x01\n\rEmbedComplete\x12\x11\n\tembedding\x18\x01 \x03(\x02\x12\x15\n\rprompt_tokens\x18\x02 \x01(\x05\x12\x15\n\rcached_tokens\x18\x03 \x01(\x05\x12\x15\n\rembedding_dim\x18\x04 \x01(\x05\x12:\n\x10\x62\x61tch_embeddings\x18\x05 \x03(\x0b\x32 .sglang.grpc.scheduler.Embedding\"*\n\tEmbedding\x12\x0e\n\x06values\x18\x01 \x03(\x02\x12\r\n\x05index\x18\x02 \x01(\x05\"<\n\nEmbedError\x12\x0f\n\x07message\x18\x01 \x01(\t\x12\x0c\n\x04\x63ode\x18\x02 \x01(\t\x12\x0f\n\x07\x64\x65tails\x18\x03 \x01(\t\"N\n\x12HealthCheckRequest\x12\x38\n\ttokenized\x18\x01 \x01(\x0b\x32%.sglang.grpc.scheduler.TokenizedInput\"7\n\x13HealthCheckResponse\x12\x0f\n\x07healthy\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\"2\n\x0c\x41\x62ortRequest\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12\x0e\n\x06reason\x18\x02 \x01(\t\"1\n\rAbortResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\"I\n\x0fLoadLoRARequest\x12\x12\n\nadapter_id\x18\x01 \x01(\t\x12\x14\n\x0c\x61\x64\x61pter_path\x18\x02 \x01(\t\x12\x0c\n\x04rank\x18\x03 \x01(\x05\"H\n\x10LoadLoRAResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x12\n\nadapter_id\x18\x02 \x01(\t\x12\x0f\n\x07message\x18\x03 \x01(\t\"\'\n\x11UnloadLoRARequest\x12\x12\n\nadapter_id\x18\x01 \x01(\t\"6\n\x12UnloadLoRAResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\"w\n\x14UpdateWeightsRequest\x12\x13\n\tdisk_path\x18\x01 \x01(\tH\x00\x12\x15\n\x0btensor_data\x18\x02 \x01(\x0cH\x00\x12\x14\n\nremote_url\x18\x03 \x01(\tH\x00\x12\x13\n\x0bweight_name\x18\x04 \x01(\tB\x08\n\x06source\"9\n\x15UpdateWeightsResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\"-\n\x17GetInternalStateRequest\x12\x12\n\nstate_keys\x18\x01 \x03(\t\"B\n\x18GetInternalStateResponse\x12&\n\x05state\x18\x01 \x01(\x0b\x32\x17.google.protobuf.Struct\"A\n\x17SetInternalStateRequest\x12&\n\x05state\x18\x01 \x01(\x0b\x32\x17.google.protobuf.Struct\"<\n\x18SetInternalStateResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\"\x15\n\x13GetModelInfoRequest\"\xea\x02\n\x14GetModelInfoResponse\x12\x12\n\nmodel_path\x18\x01 \x01(\t\x12\x16\n\x0etokenizer_path\x18\x02 \x01(\t\x12\x15\n\ris_generation\x18\x03 \x01(\x08\x12!\n\x19preferred_sampling_params\x18\x04 \x01(\t\x12\x16\n\x0eweight_version\x18\x05 \x01(\t\x12\x19\n\x11served_model_name\x18\x06 \x01(\t\x12\x1a\n\x12max_context_length\x18\x07 \x01(\x05\x12\x12\n\nvocab_size\x18\x08 \x01(\x05\x12\x17\n\x0fsupports_vision\x18\t \x01(\x08\x12\x12\n\nmodel_type\x18\n \x01(\t\x12\x15\n\reos_token_ids\x18\x0b \x03(\x05\x12\x14\n\x0cpad_token_id\x18\x0c \x01(\x05\x12\x14\n\x0c\x62os_token_id\x18\r \x01(\x05\x12\x19\n\x11max_req_input_len\x18\x0e \x01(\x05\"\x16\n\x14GetServerInfoRequest\"\xb7\x02\n\x15GetServerInfoResponse\x12,\n\x0bserver_args\x18\x01 \x01(\x0b\x32\x17.google.protobuf.Struct\x12/\n\x0escheduler_info\x18\x02 \x01(\x0b\x32\x17.google.protobuf.Struct\x12\x17\n\x0f\x61\x63tive_requests\x18\x03 \x01(\x05\x12\x11\n\tis_paused\x18\x04 \x01(\x08\x12\x1e\n\x16last_receive_timestamp\x18\x05 \x01(\x01\x12\x16\n\x0euptime_seconds\x18\x06 \x01(\x01\x12\x16\n\x0esglang_version\x18\x07 \x01(\t\x12\x13\n\x0bserver_type\x18\x08 \x01(\t\x12.\n\nstart_time\x18\t \x01(\x0b\x32\x1a.google.protobuf.Timestamp2\xd3\x04\n\x0fSglangScheduler\x12]\n\x08Generate\x12&.sglang.grpc.scheduler.GenerateRequest\x1a\'.sglang.grpc.scheduler.GenerateResponse0\x01\x12R\n\x05\x45mbed\x12#.sglang.grpc.scheduler.EmbedRequest\x1a$.sglang.grpc.scheduler.EmbedResponse\x12\x64\n\x0bHealthCheck\x12).sglang.grpc.scheduler.HealthCheckRequest\x1a*.sglang.grpc.scheduler.HealthCheckResponse\x12R\n\x05\x41\x62ort\x12#.sglang.grpc.scheduler.AbortRequest\x1a$.sglang.grpc.scheduler.AbortResponse\x12g\n\x0cGetModelInfo\x12*.sglang.grpc.scheduler.GetModelInfoRequest\x1a+.sglang.grpc.scheduler.GetModelInfoResponse\x12j\n\rGetServerInfo\x12+.sglang.grpc.scheduler.GetServerInfoRequest\x1a,.sglang.grpc.scheduler.GetServerInfoResponseb\x06proto3')
 
 _globals = globals()
 _builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals)
@@ -106,6 +106,14 @@ if not _descriptor._USE_C_DESCRIPTORS:
   _globals['_SETINTERNALSTATEREQUEST']._serialized_end=4875
   _globals['_SETINTERNALSTATERESPONSE']._serialized_start=4877
   _globals['_SETINTERNALSTATERESPONSE']._serialized_end=4937
-  _globals['_SGLANGSCHEDULER']._serialized_start=4940
-  _globals['_SGLANGSCHEDULER']._serialized_end=5322
+  _globals['_GETMODELINFOREQUEST']._serialized_start=4939
+  _globals['_GETMODELINFOREQUEST']._serialized_end=4960
+  _globals['_GETMODELINFORESPONSE']._serialized_start=4963
+  _globals['_GETMODELINFORESPONSE']._serialized_end=5325
+  _globals['_GETSERVERINFOREQUEST']._serialized_start=5327
+  _globals['_GETSERVERINFOREQUEST']._serialized_end=5349
+  _globals['_GETSERVERINFORESPONSE']._serialized_start=5352
+  _globals['_GETSERVERINFORESPONSE']._serialized_end=5663
+  _globals['_SGLANGSCHEDULER']._serialized_start=5666
+  _globals['_SGLANGSCHEDULER']._serialized_end=6261
 # @@protoc_insertion_point(module_scope)

python/sglang/srt/grpc/sglang_scheduler_pb2.pyi

@@ -428,3 +428,65 @@ class SetInternalStateResponse(_message.Message):
     success: bool
     message: str
     def __init__(self, success: bool = ..., message: _Optional[str] = ...) -> None: ...
+
+class GetModelInfoRequest(_message.Message):
+    __slots__ = ()
+    def __init__(self) -> None: ...
+
+class GetModelInfoResponse(_message.Message):
+    __slots__ = ("model_path", "tokenizer_path", "is_generation", "preferred_sampling_params", "weight_version", "served_model_name", "max_context_length", "vocab_size", "supports_vision", "model_type", "eos_token_ids", "pad_token_id", "bos_token_id", "max_req_input_len")
+    MODEL_PATH_FIELD_NUMBER: _ClassVar[int]
+    TOKENIZER_PATH_FIELD_NUMBER: _ClassVar[int]
+    IS_GENERATION_FIELD_NUMBER: _ClassVar[int]
+    PREFERRED_SAMPLING_PARAMS_FIELD_NUMBER: _ClassVar[int]
+    WEIGHT_VERSION_FIELD_NUMBER: _ClassVar[int]
+    SERVED_MODEL_NAME_FIELD_NUMBER: _ClassVar[int]
+    MAX_CONTEXT_LENGTH_FIELD_NUMBER: _ClassVar[int]
+    VOCAB_SIZE_FIELD_NUMBER: _ClassVar[int]
+    SUPPORTS_VISION_FIELD_NUMBER: _ClassVar[int]
+    MODEL_TYPE_FIELD_NUMBER: _ClassVar[int]
+    EOS_TOKEN_IDS_FIELD_NUMBER: _ClassVar[int]
+    PAD_TOKEN_ID_FIELD_NUMBER: _ClassVar[int]
+    BOS_TOKEN_ID_FIELD_NUMBER: _ClassVar[int]
+    MAX_REQ_INPUT_LEN_FIELD_NUMBER: _ClassVar[int]
+    model_path: str
+    tokenizer_path: str
+    is_generation: bool
+    preferred_sampling_params: str
+    weight_version: str
+    served_model_name: str
+    max_context_length: int
+    vocab_size: int
+    supports_vision: bool
+    model_type: str
+    eos_token_ids: _containers.RepeatedScalarFieldContainer[int]
+    pad_token_id: int
+    bos_token_id: int
+    max_req_input_len: int
+    def __init__(self, model_path: _Optional[str] = ..., tokenizer_path: _Optional[str] = ..., is_generation: bool = ..., preferred_sampling_params: _Optional[str] = ..., weight_version: _Optional[str] = ..., served_model_name: _Optional[str] = ..., max_context_length: _Optional[int] = ..., vocab_size: _Optional[int] = ..., supports_vision: bool = ..., model_type: _Optional[str] = ..., eos_token_ids: _Optional[_Iterable[int]] = ..., pad_token_id: _Optional[int] = ..., bos_token_id: _Optional[int] = ..., max_req_input_len: _Optional[int] = ...) -> None: ...
+
+class GetServerInfoRequest(_message.Message):
+    __slots__ = ()
+    def __init__(self) -> None: ...
+
+class GetServerInfoResponse(_message.Message):
+    __slots__ = ("server_args", "scheduler_info", "active_requests", "is_paused", "last_receive_timestamp", "uptime_seconds", "sglang_version", "server_type", "start_time")
+    SERVER_ARGS_FIELD_NUMBER: _ClassVar[int]
+    SCHEDULER_INFO_FIELD_NUMBER: _ClassVar[int]
+    ACTIVE_REQUESTS_FIELD_NUMBER: _ClassVar[int]
+    IS_PAUSED_FIELD_NUMBER: _ClassVar[int]
+    LAST_RECEIVE_TIMESTAMP_FIELD_NUMBER: _ClassVar[int]
+    UPTIME_SECONDS_FIELD_NUMBER: _ClassVar[int]
+    SGLANG_VERSION_FIELD_NUMBER: _ClassVar[int]
+    SERVER_TYPE_FIELD_NUMBER: _ClassVar[int]
+    START_TIME_FIELD_NUMBER: _ClassVar[int]
+    server_args: _struct_pb2.Struct
+    scheduler_info: _struct_pb2.Struct
+    active_requests: int
+    is_paused: bool
+    last_receive_timestamp: float
+    uptime_seconds: float
+    sglang_version: str
+    server_type: str
+    start_time: _timestamp_pb2.Timestamp
+    def __init__(self, server_args: _Optional[_Union[_struct_pb2.Struct, _Mapping]] = ..., scheduler_info: _Optional[_Union[_struct_pb2.Struct, _Mapping]] = ..., active_requests: _Optional[int] = ..., is_paused: bool = ..., last_receive_timestamp: _Optional[float] = ..., uptime_seconds: _Optional[float] = ..., sglang_version: _Optional[str] = ..., server_type: _Optional[str] = ..., start_time: _Optional[_Union[datetime.datetime, _timestamp_pb2.Timestamp, _Mapping]] = ...) -> None: ...

python/sglang/srt/grpc/sglang_scheduler_pb2_grpc.py

@@ -59,6 +59,16 @@ class SglangSchedulerStub(object):
                 request_serializer=sglang__scheduler__pb2.AbortRequest.SerializeToString,
                 response_deserializer=sglang__scheduler__pb2.AbortResponse.FromString,
                 _registered_method=True)
+        self.GetModelInfo = channel.unary_unary(
+                '/sglang.grpc.scheduler.SglangScheduler/GetModelInfo',
+                request_serializer=sglang__scheduler__pb2.GetModelInfoRequest.SerializeToString,
+                response_deserializer=sglang__scheduler__pb2.GetModelInfoResponse.FromString,
+                _registered_method=True)
+        self.GetServerInfo = channel.unary_unary(
+                '/sglang.grpc.scheduler.SglangScheduler/GetServerInfo',
+                request_serializer=sglang__scheduler__pb2.GetServerInfoRequest.SerializeToString,
+                response_deserializer=sglang__scheduler__pb2.GetServerInfoResponse.FromString,
+                _registered_method=True)
 
 
 class SglangSchedulerServicer(object):
@@ -94,6 +104,20 @@ class SglangSchedulerServicer(object):
         context.set_details('Method not implemented!')
         raise NotImplementedError('Method not implemented!')
 
+    def GetModelInfo(self, request, context):
+        """Get model information
+        """
+        context.set_code(grpc.StatusCode.UNIMPLEMENTED)
+        context.set_details('Method not implemented!')
+        raise NotImplementedError('Method not implemented!')
+
+    def GetServerInfo(self, request, context):
+        """Get server information
+        """
+        context.set_code(grpc.StatusCode.UNIMPLEMENTED)
+        context.set_details('Method not implemented!')
+        raise NotImplementedError('Method not implemented!')
+
 
 def add_SglangSchedulerServicer_to_server(servicer, server):
     rpc_method_handlers = {
@@ -117,6 +141,16 @@ def add_SglangSchedulerServicer_to_server(servicer, server):
                     request_deserializer=sglang__scheduler__pb2.AbortRequest.FromString,
                     response_serializer=sglang__scheduler__pb2.AbortResponse.SerializeToString,
             ),
+            'GetModelInfo': grpc.unary_unary_rpc_method_handler(
+                    servicer.GetModelInfo,
+                    request_deserializer=sglang__scheduler__pb2.GetModelInfoRequest.FromString,
+                    response_serializer=sglang__scheduler__pb2.GetModelInfoResponse.SerializeToString,
+            ),
+            'GetServerInfo': grpc.unary_unary_rpc_method_handler(
+                    servicer.GetServerInfo,
+                    request_deserializer=sglang__scheduler__pb2.GetServerInfoRequest.FromString,
+                    response_serializer=sglang__scheduler__pb2.GetServerInfoResponse.SerializeToString,
+            ),
     }
     generic_handler = grpc.method_handlers_generic_handler(
             'sglang.grpc.scheduler.SglangScheduler', rpc_method_handlers)
@@ -237,3 +271,57 @@ class SglangScheduler(object):
             timeout,
             metadata,
             _registered_method=True)
+
+    @staticmethod
+    def GetModelInfo(request,
+            target,
+            options=(),
+            channel_credentials=None,
+            call_credentials=None,
+            insecure=False,
+            compression=None,
+            wait_for_ready=None,
+            timeout=None,
+            metadata=None):
+        return grpc.experimental.unary_unary(
+            request,
+            target,
+            '/sglang.grpc.scheduler.SglangScheduler/GetModelInfo',
+            sglang__scheduler__pb2.GetModelInfoRequest.SerializeToString,
+            sglang__scheduler__pb2.GetModelInfoResponse.FromString,
+            options,
+            channel_credentials,
+            insecure,
+            call_credentials,
+            compression,
+            wait_for_ready,
+            timeout,
+            metadata,
+            _registered_method=True)
+
+    @staticmethod
+    def GetServerInfo(request,
+            target,
+            options=(),
+            channel_credentials=None,
+            call_credentials=None,
+            insecure=False,
+            compression=None,
+            wait_for_ready=None,
+            timeout=None,
+            metadata=None):
+        return grpc.experimental.unary_unary(
+            request,
+            target,
+            '/sglang.grpc.scheduler.SglangScheduler/GetServerInfo',
+            sglang__scheduler__pb2.GetServerInfoRequest.SerializeToString,
+            sglang__scheduler__pb2.GetServerInfoResponse.FromString,
+            options,
+            channel_credentials,
+            insecure,
+            call_credentials,
+            compression,
+            wait_for_ready,
+            timeout,
+            metadata,
+            _registered_method=True)

sgl-router/src/grpc_client/sglang_scheduler.rs

@@ -97,6 +97,30 @@ impl SglangSchedulerClient {
         Ok(())
     }
 
+    /// Get model information
+    pub async fn get_model_info(
+        &mut self,
+    ) -> Result<proto::GetModelInfoResponse, Box<dyn std::error::Error + Send + Sync>> {
+        debug!("Requesting model info");
+        let request = Request::new(proto::GetModelInfoRequest {});
+
+        let response = self.client.get_model_info(request).await?;
+        debug!("Model info response received");
+        Ok(response.into_inner())
+    }
+
+    /// Get server information
+    pub async fn get_server_info(
+        &mut self,
+    ) -> Result<proto::GetServerInfoResponse, Box<dyn std::error::Error + Send + Sync>> {
+        debug!("Requesting server info");
+        let request = Request::new(proto::GetServerInfoRequest {});
+
+        let response = self.client.get_server_info(request).await?;
+        debug!("Server info response received");
+        Ok(response.into_inner())
+    }
+
     /// Build a single SGLang GenerateRequest from OpenAI ChatCompletionRequest
     pub fn build_generate_request(
         &self,

sgl-router/src/proto/sglang_scheduler.proto

@@ -20,6 +20,12 @@ service SglangScheduler {
   // Abort a running request
   rpc Abort(AbortRequest) returns (AbortResponse);
 
+  // Get model information
+  rpc GetModelInfo(GetModelInfoRequest) returns (GetModelInfoResponse);
+
+  // Get server information
+  rpc GetServerInfo(GetServerInfoRequest) returns (GetServerInfoResponse);
+
 }
 
 // =====================
@@ -401,3 +407,56 @@ message SetInternalStateResponse {
   bool success = 1;
   string message = 2;
 }
+
+// =====================
+// Model and Server Info
+// =====================
+
+// Get model information
+message GetModelInfoRequest {}
+
+message GetModelInfoResponse {
+  string model_path = 1;
+  string tokenizer_path = 2;
+  bool is_generation = 3;
+  string preferred_sampling_params = 4;  // JSON string or empty
+  string weight_version = 5;
+  string served_model_name = 6;
+  int32 max_context_length = 7;
+  int32 vocab_size = 8;
+  bool supports_vision = 9;
+  string model_type = 10;
+  repeated int32 eos_token_ids = 11;
+  int32 pad_token_id = 12;
+  int32 bos_token_id = 13;
+  int32 max_req_input_len = 14;
+}
+
+// Get server information
+message GetServerInfoRequest {}
+
+message GetServerInfoResponse {
+  // Server configuration (as structured data)
+  google.protobuf.Struct server_args = 1;
+
+  // Scheduler metrics (from scheduler initialization)
+  google.protobuf.Struct scheduler_info = 2;
+
+  // Runtime state
+  int32 active_requests = 3;
+  bool is_paused = 4;
+  double last_receive_timestamp = 5;
+  double uptime_seconds = 6;
+
+  // Version info
+  string sglang_version = 7;
+
+  // Server metadata
+  string server_type = 8;  // "grpc"
+  google.protobuf.Timestamp start_time = 9;
+
+  // Note: internal_states not provided in gRPC mode
+  // Scheduler-side metrics (memory usage, throughput) require
+  // bidirectional communicator infrastructure not available in gRPC.
+  // Use HTTP /get_server_info if scheduler internal state is needed.
+}