From 7dcd689b477ec1a274f8f833ba88a770e441a061 Mon Sep 17 00:00:00 2001 From: Chang Su Date: Thu, 25 Sep 2025 17:48:47 -0700 Subject: [PATCH] [router][refactor] Clean up protobuf fields (#10923) --- python/sglang/srt/entrypoints/grpc_server.py | 15 +-- python/sglang/srt/grpc/sglang_scheduler.proto | 30 +++-- .../sglang/srt/grpc/sglang_scheduler_pb2.py | 108 +++++++++--------- .../sglang/srt/grpc/sglang_scheduler_pb2.pyi | 28 ++--- sgl-router/src/proto/sglang_scheduler.proto | 30 +++-- 5 files changed, 102 insertions(+), 109 deletions(-) diff --git a/python/sglang/srt/entrypoints/grpc_server.py b/python/sglang/srt/entrypoints/grpc_server.py index 5e91210f4..fa1b1143d 100644 --- a/python/sglang/srt/entrypoints/grpc_server.py +++ b/python/sglang/srt/entrypoints/grpc_server.py @@ -266,7 +266,6 @@ class SGLangSchedulerServicer(sglang_scheduler_pb2_grpc.SglangSchedulerServicer) prompt_tokens=result.get("prompt_tokens", 0), cached_tokens=0, embedding_dim=len(result["embedding"]), - generation_time=time.time() - self.start_time, ), ) @@ -477,16 +476,14 @@ class SGLangSchedulerServicer(sglang_scheduler_pb2_grpc.SglangSchedulerServicer) self, request_id: str, output: Dict ) -> sglang_scheduler_pb2.GenerateResponse: """Create a streaming chunk response.""" + meta_info = output.get("meta_info", {}) return sglang_scheduler_pb2.GenerateResponse( request_id=request_id, chunk=sglang_scheduler_pb2.GenerateStreamChunk( token_id=output["token_ids"][-1] if output.get("token_ids") else 0, - text=output.get("text", ""), - prompt_tokens=0, - completion_tokens=len(output.get("token_ids", [])), + prompt_tokens=meta_info.get("prompt_tokens", 0), + completion_tokens=meta_info.get("completion_tokens", 0), cached_tokens=0, - generation_time=time.time() - self.start_time, - queue_time=0.0, ), ) @@ -507,8 +504,12 @@ class SGLangSchedulerServicer(sglang_scheduler_pb2_grpc.SglangSchedulerServicer) request_id=request_id, complete=sglang_scheduler_pb2.GenerateComplete( output_ids=output.get("token_ids", []), - output_text=output.get("text", ""), finish_reason=finish_reason, + prompt_tokens=meta_info.get("prompt_tokens", 0), + completion_tokens=meta_info.get( + "completion_tokens", len(output.get("token_ids", [])) + ), + cached_tokens=meta_info.get("cached_tokens", 0), ), ) diff --git a/python/sglang/srt/grpc/sglang_scheduler.proto b/python/sglang/srt/grpc/sglang_scheduler.proto index affa131db..e4638e7a9 100644 --- a/python/sglang/srt/grpc/sglang_scheduler.proto +++ b/python/sglang/srt/grpc/sglang_scheduler.proto @@ -165,28 +165,22 @@ message GenerateResponse { message GenerateStreamChunk { // Generated token int32 token_id = 1; - string text = 2; // Cumulative counts - int32 prompt_tokens = 3; - int32 completion_tokens = 4; - int32 cached_tokens = 5; + int32 prompt_tokens = 2; + int32 completion_tokens = 3; + int32 cached_tokens = 4; // Logprobs (if requested) - LogProbs logprobs = 6; + LogProbs logprobs = 5; // Hidden states (if requested) - repeated float hidden_states = 7; - - // Metadata - float generation_time = 8; // Time to generate this token - int32 queue_time = 9; // Time spent in queue + repeated float hidden_states = 6; } message GenerateComplete { // Final output repeated int32 output_ids = 1; - string output_text = 2; // Finish reason enum FinishReason { @@ -201,13 +195,18 @@ message GenerateComplete { // The request was aborted by the user or system. ABORT = 4; } - FinishReason finish_reason = 3; + FinishReason finish_reason = 2; + + // Token usage counts + int32 prompt_tokens = 3; + int32 completion_tokens = 4; + int32 cached_tokens = 5; // All logprobs if requested - repeated LogProbs all_logprobs = 11; + repeated LogProbs all_logprobs = 6; // All hidden states if requested - repeated HiddenStates all_hidden_states = 12; + repeated HiddenStates all_hidden_states = 7; } message GenerateError { @@ -285,10 +284,9 @@ message EmbedComplete { // Additional metadata int32 embedding_dim = 4; - float generation_time = 5; // For batch embeddings - repeated Embedding batch_embeddings = 6; + repeated Embedding batch_embeddings = 5; } message Embedding { diff --git a/python/sglang/srt/grpc/sglang_scheduler_pb2.py b/python/sglang/srt/grpc/sglang_scheduler_pb2.py index 2f614132a..8b05bf3fc 100644 --- a/python/sglang/srt/grpc/sglang_scheduler_pb2.py +++ b/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\"\xc9\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\x16\n\x0emax_new_tokens\x18\x08 \x01(\x05\x12\x0c\n\x04stop\x18\t \x03(\t\x12\x16\n\x0estop_token_ids\x18\n \x03(\x05\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\x11\n\tlora_path\x18\x11 \x01(\t\x12\t\n\x01n\x18\x12 \x01(\x05\x12\x15\n\rtoken_healing\x18\x13 \x01(\x08\x12\x16\n\x0emin_new_tokens\x18\x14 \x01(\x05\x12\x12\n\nignore_eos\x18\x15 \x01(\x08\x12\x14\n\x0cno_stop_trim\x18\x16 \x01(\x08\x12\x17\n\x0fstream_interval\x18\x17 \x01(\x05\x12H\n\nlogit_bias\x18\x18 \x03(\x0b\x32\x34.sglang.grpc.scheduler.SamplingParams.LogitBiasEntry\x12.\n\rcustom_params\x18\x19 \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\nconstraint\"]\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\"\xe9\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(\x05\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\x15\n\rdp_balance_id\x18\x11 \x01(\x05\":\n\x0eTokenizedInput\x12\x15\n\roriginal_text\x18\x01 \x01(\t\x12\x11\n\tinput_ids\x18\x02 \x03(\x05\"\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\"\xf5\x01\n\x13GenerateStreamChunk\x12\x10\n\x08token_id\x18\x01 \x01(\x05\x12\x0c\n\x04text\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\x31\n\x08logprobs\x18\x06 \x01(\x0b\x32\x1f.sglang.grpc.scheduler.LogProbs\x12\x15\n\rhidden_states\x18\x07 \x03(\x02\x12\x17\n\x0fgeneration_time\x18\x08 \x01(\x02\x12\x12\n\nqueue_time\x18\t \x01(\x05\"\xcd\x02\n\x10GenerateComplete\x12\x12\n\noutput_ids\x18\x01 \x03(\x05\x12\x13\n\x0boutput_text\x18\x02 \x01(\t\x12K\n\rfinish_reason\x18\x03 \x01(\x0e\x32\x34.sglang.grpc.scheduler.GenerateComplete.FinishReason\x12\x35\n\x0c\x61ll_logprobs\x18\x0b \x03(\x0b\x32\x1f.sglang.grpc.scheduler.LogProbs\x12>\n\x11\x61ll_hidden_states\x18\x0c \x03(\x0b\x32#.sglang.grpc.scheduler.HiddenStates\"L\n\x0c\x46inishReason\x12\x08\n\x04STOP\x10\x00\x12\n\n\x06LENGTH\x10\x01\x12\r\n\tEOS_TOKEN\x10\x02\x12\x0c\n\x08STOP_STR\x10\x03\x12\t\n\x05\x41\x42ORT\x10\x04\"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\"\x84\x01\n\x08LogProbs\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\x12\x13\n\x0btoken_texts\x18\x04 \x03(\t\"E\n\x0bTopLogProbs\x12\x0e\n\x06values\x18\x01 \x03(\x02\x12\x11\n\ttoken_ids\x18\x02 \x03(\x05\x12\x13\n\x0btoken_texts\x18\x03 \x03(\t\"?\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\"\xbc\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\x17\n\x0fgeneration_time\x18\x05 \x01(\x02\x12:\n\x10\x62\x61tch_embeddings\x18\x06 \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\"\xc9\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\x16\n\x0emax_new_tokens\x18\x08 \x01(\x05\x12\x0c\n\x04stop\x18\t \x03(\t\x12\x16\n\x0estop_token_ids\x18\n \x03(\x05\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\x11\n\tlora_path\x18\x11 \x01(\t\x12\t\n\x01n\x18\x12 \x01(\x05\x12\x15\n\rtoken_healing\x18\x13 \x01(\x08\x12\x16\n\x0emin_new_tokens\x18\x14 \x01(\x05\x12\x12\n\nignore_eos\x18\x15 \x01(\x08\x12\x14\n\x0cno_stop_trim\x18\x16 \x01(\x08\x12\x17\n\x0fstream_interval\x18\x17 \x01(\x05\x12H\n\nlogit_bias\x18\x18 \x03(\x0b\x32\x34.sglang.grpc.scheduler.SamplingParams.LogitBiasEntry\x12.\n\rcustom_params\x18\x19 \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\nconstraint\"]\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\"\xe9\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(\x05\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\x15\n\rdp_balance_id\x18\x11 \x01(\x05\":\n\x0eTokenizedInput\x12\x15\n\roriginal_text\x18\x01 \x01(\t\x12\x11\n\tinput_ids\x18\x02 \x03(\x05\"\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\"\xba\x01\n\x13GenerateStreamChunk\x12\x10\n\x08token_id\x18\x01 \x01(\x05\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\x31\n\x08logprobs\x18\x05 \x01(\x0b\x32\x1f.sglang.grpc.scheduler.LogProbs\x12\x15\n\rhidden_states\x18\x06 \x03(\x02\"\x81\x03\n\x10GenerateComplete\x12\x12\n\noutput_ids\x18\x01 \x03(\x05\x12K\n\rfinish_reason\x18\x02 \x01(\x0e\x32\x34.sglang.grpc.scheduler.GenerateComplete.FinishReason\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\x35\n\x0c\x61ll_logprobs\x18\x06 \x03(\x0b\x32\x1f.sglang.grpc.scheduler.LogProbs\x12>\n\x11\x61ll_hidden_states\x18\x07 \x03(\x0b\x32#.sglang.grpc.scheduler.HiddenStates\"L\n\x0c\x46inishReason\x12\x08\n\x04STOP\x10\x00\x12\n\n\x06LENGTH\x10\x01\x12\r\n\tEOS_TOKEN\x10\x02\x12\x0c\n\x08STOP_STR\x10\x03\x12\t\n\x05\x41\x42ORT\x10\x04\"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\"\x84\x01\n\x08LogProbs\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\x12\x13\n\x0btoken_texts\x18\x04 \x03(\t\"E\n\x0bTopLogProbs\x12\x0e\n\x06values\x18\x01 \x03(\x02\x12\x11\n\ttoken_ids\x18\x02 \x03(\x05\x12\x13\n\x0btoken_texts\x18\x03 \x03(\t\"?\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') _globals = globals() _builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals) @@ -53,57 +53,57 @@ if not _descriptor._USE_C_DESCRIPTORS: _globals['_GENERATERESPONSE']._serialized_start=1818 _globals['_GENERATERESPONSE']._serialized_end=2045 _globals['_GENERATESTREAMCHUNK']._serialized_start=2048 - _globals['_GENERATESTREAMCHUNK']._serialized_end=2293 - _globals['_GENERATECOMPLETE']._serialized_start=2296 - _globals['_GENERATECOMPLETE']._serialized_end=2629 - _globals['_GENERATECOMPLETE_FINISHREASON']._serialized_start=2553 - _globals['_GENERATECOMPLETE_FINISHREASON']._serialized_end=2629 - _globals['_GENERATEERROR']._serialized_start=2631 - _globals['_GENERATEERROR']._serialized_end=2706 - _globals['_LOGPROBS']._serialized_start=2709 - _globals['_LOGPROBS']._serialized_end=2841 - _globals['_TOPLOGPROBS']._serialized_start=2843 - _globals['_TOPLOGPROBS']._serialized_end=2912 - _globals['_HIDDENSTATES']._serialized_start=2914 - _globals['_HIDDENSTATES']._serialized_end=2977 - _globals['_EMBEDREQUEST']._serialized_start=2980 - _globals['_EMBEDREQUEST']._serialized_end=3310 - _globals['_EMBEDRESPONSE']._serialized_start=3313 - _globals['_EMBEDRESPONSE']._serialized_end=3470 - _globals['_EMBEDCOMPLETE']._serialized_start=3473 - _globals['_EMBEDCOMPLETE']._serialized_end=3661 - _globals['_EMBEDDING']._serialized_start=3663 - _globals['_EMBEDDING']._serialized_end=3705 - _globals['_EMBEDERROR']._serialized_start=3707 - _globals['_EMBEDERROR']._serialized_end=3767 - _globals['_HEALTHCHECKREQUEST']._serialized_start=3769 - _globals['_HEALTHCHECKREQUEST']._serialized_end=3847 - _globals['_HEALTHCHECKRESPONSE']._serialized_start=3849 - _globals['_HEALTHCHECKRESPONSE']._serialized_end=3904 - _globals['_ABORTREQUEST']._serialized_start=3906 - _globals['_ABORTREQUEST']._serialized_end=3956 - _globals['_ABORTRESPONSE']._serialized_start=3958 - _globals['_ABORTRESPONSE']._serialized_end=4007 - _globals['_LOADLORAREQUEST']._serialized_start=4009 - _globals['_LOADLORAREQUEST']._serialized_end=4082 - _globals['_LOADLORARESPONSE']._serialized_start=4084 - _globals['_LOADLORARESPONSE']._serialized_end=4156 - _globals['_UNLOADLORAREQUEST']._serialized_start=4158 - _globals['_UNLOADLORAREQUEST']._serialized_end=4197 - _globals['_UNLOADLORARESPONSE']._serialized_start=4199 - _globals['_UNLOADLORARESPONSE']._serialized_end=4253 - _globals['_UPDATEWEIGHTSREQUEST']._serialized_start=4255 - _globals['_UPDATEWEIGHTSREQUEST']._serialized_end=4374 - _globals['_UPDATEWEIGHTSRESPONSE']._serialized_start=4376 - _globals['_UPDATEWEIGHTSRESPONSE']._serialized_end=4433 - _globals['_GETINTERNALSTATEREQUEST']._serialized_start=4435 - _globals['_GETINTERNALSTATEREQUEST']._serialized_end=4480 - _globals['_GETINTERNALSTATERESPONSE']._serialized_start=4482 - _globals['_GETINTERNALSTATERESPONSE']._serialized_end=4548 - _globals['_SETINTERNALSTATEREQUEST']._serialized_start=4550 - _globals['_SETINTERNALSTATEREQUEST']._serialized_end=4615 - _globals['_SETINTERNALSTATERESPONSE']._serialized_start=4617 - _globals['_SETINTERNALSTATERESPONSE']._serialized_end=4677 - _globals['_SGLANGSCHEDULER']._serialized_start=4680 - _globals['_SGLANGSCHEDULER']._serialized_end=5062 + _globals['_GENERATESTREAMCHUNK']._serialized_end=2234 + _globals['_GENERATECOMPLETE']._serialized_start=2237 + _globals['_GENERATECOMPLETE']._serialized_end=2622 + _globals['_GENERATECOMPLETE_FINISHREASON']._serialized_start=2546 + _globals['_GENERATECOMPLETE_FINISHREASON']._serialized_end=2622 + _globals['_GENERATEERROR']._serialized_start=2624 + _globals['_GENERATEERROR']._serialized_end=2699 + _globals['_LOGPROBS']._serialized_start=2702 + _globals['_LOGPROBS']._serialized_end=2834 + _globals['_TOPLOGPROBS']._serialized_start=2836 + _globals['_TOPLOGPROBS']._serialized_end=2905 + _globals['_HIDDENSTATES']._serialized_start=2907 + _globals['_HIDDENSTATES']._serialized_end=2970 + _globals['_EMBEDREQUEST']._serialized_start=2973 + _globals['_EMBEDREQUEST']._serialized_end=3303 + _globals['_EMBEDRESPONSE']._serialized_start=3306 + _globals['_EMBEDRESPONSE']._serialized_end=3463 + _globals['_EMBEDCOMPLETE']._serialized_start=3466 + _globals['_EMBEDCOMPLETE']._serialized_end=3629 + _globals['_EMBEDDING']._serialized_start=3631 + _globals['_EMBEDDING']._serialized_end=3673 + _globals['_EMBEDERROR']._serialized_start=3675 + _globals['_EMBEDERROR']._serialized_end=3735 + _globals['_HEALTHCHECKREQUEST']._serialized_start=3737 + _globals['_HEALTHCHECKREQUEST']._serialized_end=3815 + _globals['_HEALTHCHECKRESPONSE']._serialized_start=3817 + _globals['_HEALTHCHECKRESPONSE']._serialized_end=3872 + _globals['_ABORTREQUEST']._serialized_start=3874 + _globals['_ABORTREQUEST']._serialized_end=3924 + _globals['_ABORTRESPONSE']._serialized_start=3926 + _globals['_ABORTRESPONSE']._serialized_end=3975 + _globals['_LOADLORAREQUEST']._serialized_start=3977 + _globals['_LOADLORAREQUEST']._serialized_end=4050 + _globals['_LOADLORARESPONSE']._serialized_start=4052 + _globals['_LOADLORARESPONSE']._serialized_end=4124 + _globals['_UNLOADLORAREQUEST']._serialized_start=4126 + _globals['_UNLOADLORAREQUEST']._serialized_end=4165 + _globals['_UNLOADLORARESPONSE']._serialized_start=4167 + _globals['_UNLOADLORARESPONSE']._serialized_end=4221 + _globals['_UPDATEWEIGHTSREQUEST']._serialized_start=4223 + _globals['_UPDATEWEIGHTSREQUEST']._serialized_end=4342 + _globals['_UPDATEWEIGHTSRESPONSE']._serialized_start=4344 + _globals['_UPDATEWEIGHTSRESPONSE']._serialized_end=4401 + _globals['_GETINTERNALSTATEREQUEST']._serialized_start=4403 + _globals['_GETINTERNALSTATEREQUEST']._serialized_end=4448 + _globals['_GETINTERNALSTATERESPONSE']._serialized_start=4450 + _globals['_GETINTERNALSTATERESPONSE']._serialized_end=4516 + _globals['_SETINTERNALSTATEREQUEST']._serialized_start=4518 + _globals['_SETINTERNALSTATEREQUEST']._serialized_end=4583 + _globals['_SETINTERNALSTATERESPONSE']._serialized_start=4585 + _globals['_SETINTERNALSTATERESPONSE']._serialized_end=4645 + _globals['_SGLANGSCHEDULER']._serialized_start=4648 + _globals['_SGLANGSCHEDULER']._serialized_end=5030 # @@protoc_insertion_point(module_scope) diff --git a/python/sglang/srt/grpc/sglang_scheduler_pb2.pyi b/python/sglang/srt/grpc/sglang_scheduler_pb2.pyi index 34d9cd0fe..bf383f127 100644 --- a/python/sglang/srt/grpc/sglang_scheduler_pb2.pyi +++ b/python/sglang/srt/grpc/sglang_scheduler_pb2.pyi @@ -161,29 +161,23 @@ class GenerateResponse(_message.Message): def __init__(self, request_id: _Optional[str] = ..., chunk: _Optional[_Union[GenerateStreamChunk, _Mapping]] = ..., complete: _Optional[_Union[GenerateComplete, _Mapping]] = ..., error: _Optional[_Union[GenerateError, _Mapping]] = ...) -> None: ... class GenerateStreamChunk(_message.Message): - __slots__ = ("token_id", "text", "prompt_tokens", "completion_tokens", "cached_tokens", "logprobs", "hidden_states", "generation_time", "queue_time") + __slots__ = ("token_id", "prompt_tokens", "completion_tokens", "cached_tokens", "logprobs", "hidden_states") TOKEN_ID_FIELD_NUMBER: _ClassVar[int] - TEXT_FIELD_NUMBER: _ClassVar[int] PROMPT_TOKENS_FIELD_NUMBER: _ClassVar[int] COMPLETION_TOKENS_FIELD_NUMBER: _ClassVar[int] CACHED_TOKENS_FIELD_NUMBER: _ClassVar[int] LOGPROBS_FIELD_NUMBER: _ClassVar[int] HIDDEN_STATES_FIELD_NUMBER: _ClassVar[int] - GENERATION_TIME_FIELD_NUMBER: _ClassVar[int] - QUEUE_TIME_FIELD_NUMBER: _ClassVar[int] token_id: int - text: str prompt_tokens: int completion_tokens: int cached_tokens: int logprobs: LogProbs hidden_states: _containers.RepeatedScalarFieldContainer[float] - generation_time: float - queue_time: int - def __init__(self, token_id: _Optional[int] = ..., text: _Optional[str] = ..., prompt_tokens: _Optional[int] = ..., completion_tokens: _Optional[int] = ..., cached_tokens: _Optional[int] = ..., logprobs: _Optional[_Union[LogProbs, _Mapping]] = ..., hidden_states: _Optional[_Iterable[float]] = ..., generation_time: _Optional[float] = ..., queue_time: _Optional[int] = ...) -> None: ... + def __init__(self, token_id: _Optional[int] = ..., prompt_tokens: _Optional[int] = ..., completion_tokens: _Optional[int] = ..., cached_tokens: _Optional[int] = ..., logprobs: _Optional[_Union[LogProbs, _Mapping]] = ..., hidden_states: _Optional[_Iterable[float]] = ...) -> None: ... class GenerateComplete(_message.Message): - __slots__ = ("output_ids", "output_text", "finish_reason", "all_logprobs", "all_hidden_states") + __slots__ = ("output_ids", "finish_reason", "prompt_tokens", "completion_tokens", "cached_tokens", "all_logprobs", "all_hidden_states") class FinishReason(int, metaclass=_enum_type_wrapper.EnumTypeWrapper): __slots__ = () STOP: _ClassVar[GenerateComplete.FinishReason] @@ -197,16 +191,20 @@ class GenerateComplete(_message.Message): STOP_STR: GenerateComplete.FinishReason ABORT: GenerateComplete.FinishReason OUTPUT_IDS_FIELD_NUMBER: _ClassVar[int] - OUTPUT_TEXT_FIELD_NUMBER: _ClassVar[int] FINISH_REASON_FIELD_NUMBER: _ClassVar[int] + PROMPT_TOKENS_FIELD_NUMBER: _ClassVar[int] + COMPLETION_TOKENS_FIELD_NUMBER: _ClassVar[int] + CACHED_TOKENS_FIELD_NUMBER: _ClassVar[int] ALL_LOGPROBS_FIELD_NUMBER: _ClassVar[int] ALL_HIDDEN_STATES_FIELD_NUMBER: _ClassVar[int] output_ids: _containers.RepeatedScalarFieldContainer[int] - output_text: str finish_reason: GenerateComplete.FinishReason + prompt_tokens: int + completion_tokens: int + cached_tokens: int all_logprobs: _containers.RepeatedCompositeFieldContainer[LogProbs] all_hidden_states: _containers.RepeatedCompositeFieldContainer[HiddenStates] - def __init__(self, output_ids: _Optional[_Iterable[int]] = ..., output_text: _Optional[str] = ..., finish_reason: _Optional[_Union[GenerateComplete.FinishReason, str]] = ..., all_logprobs: _Optional[_Iterable[_Union[LogProbs, _Mapping]]] = ..., all_hidden_states: _Optional[_Iterable[_Union[HiddenStates, _Mapping]]] = ...) -> None: ... + def __init__(self, output_ids: _Optional[_Iterable[int]] = ..., finish_reason: _Optional[_Union[GenerateComplete.FinishReason, str]] = ..., prompt_tokens: _Optional[int] = ..., completion_tokens: _Optional[int] = ..., cached_tokens: _Optional[int] = ..., all_logprobs: _Optional[_Iterable[_Union[LogProbs, _Mapping]]] = ..., all_hidden_states: _Optional[_Iterable[_Union[HiddenStates, _Mapping]]] = ...) -> None: ... class GenerateError(_message.Message): __slots__ = ("message", "http_status_code", "details") @@ -283,20 +281,18 @@ class EmbedResponse(_message.Message): def __init__(self, request_id: _Optional[str] = ..., complete: _Optional[_Union[EmbedComplete, _Mapping]] = ..., error: _Optional[_Union[EmbedError, _Mapping]] = ...) -> None: ... class EmbedComplete(_message.Message): - __slots__ = ("embedding", "prompt_tokens", "cached_tokens", "embedding_dim", "generation_time", "batch_embeddings") + __slots__ = ("embedding", "prompt_tokens", "cached_tokens", "embedding_dim", "batch_embeddings") EMBEDDING_FIELD_NUMBER: _ClassVar[int] PROMPT_TOKENS_FIELD_NUMBER: _ClassVar[int] CACHED_TOKENS_FIELD_NUMBER: _ClassVar[int] EMBEDDING_DIM_FIELD_NUMBER: _ClassVar[int] - GENERATION_TIME_FIELD_NUMBER: _ClassVar[int] BATCH_EMBEDDINGS_FIELD_NUMBER: _ClassVar[int] embedding: _containers.RepeatedScalarFieldContainer[float] prompt_tokens: int cached_tokens: int embedding_dim: int - generation_time: float batch_embeddings: _containers.RepeatedCompositeFieldContainer[Embedding] - def __init__(self, embedding: _Optional[_Iterable[float]] = ..., prompt_tokens: _Optional[int] = ..., cached_tokens: _Optional[int] = ..., embedding_dim: _Optional[int] = ..., generation_time: _Optional[float] = ..., batch_embeddings: _Optional[_Iterable[_Union[Embedding, _Mapping]]] = ...) -> None: ... + def __init__(self, embedding: _Optional[_Iterable[float]] = ..., prompt_tokens: _Optional[int] = ..., cached_tokens: _Optional[int] = ..., embedding_dim: _Optional[int] = ..., batch_embeddings: _Optional[_Iterable[_Union[Embedding, _Mapping]]] = ...) -> None: ... class Embedding(_message.Message): __slots__ = ("values", "index") diff --git a/sgl-router/src/proto/sglang_scheduler.proto b/sgl-router/src/proto/sglang_scheduler.proto index 55b3c77f8..b0e8b92c5 100644 --- a/sgl-router/src/proto/sglang_scheduler.proto +++ b/sgl-router/src/proto/sglang_scheduler.proto @@ -165,28 +165,22 @@ message GenerateResponse { message GenerateStreamChunk { // Generated token int32 token_id = 1; - string text = 2; // Cumulative counts - int32 prompt_tokens = 3; - int32 completion_tokens = 4; - int32 cached_tokens = 5; + int32 prompt_tokens = 2; + int32 completion_tokens = 3; + int32 cached_tokens = 4; // Logprobs (if requested) - LogProbs logprobs = 6; + LogProbs logprobs = 5; // Hidden states (if requested) - repeated float hidden_states = 7; - - // Metadata - float generation_time = 8; // Time to generate this token - int32 queue_time = 9; // Time spent in queue + repeated float hidden_states = 6; } message GenerateComplete { // Final output repeated int32 output_ids = 1; - string output_text = 2; // Finish reason enum FinishReason { @@ -201,13 +195,18 @@ message GenerateComplete { // The request was aborted by the user or system. ABORT = 4; } - FinishReason finish_reason = 3; + FinishReason finish_reason = 2; + + // Token usage counts + int32 prompt_tokens = 3; + int32 completion_tokens = 4; + int32 cached_tokens = 5; // All logprobs if requested - repeated LogProbs all_logprobs = 11; + repeated LogProbs all_logprobs = 6; // All hidden states if requested - repeated HiddenStates all_hidden_states = 12; + repeated HiddenStates all_hidden_states = 7; } message GenerateError { @@ -285,10 +284,9 @@ message EmbedComplete { // Additional metadata int32 embedding_dim = 4; - float generation_time = 5; // For batch embeddings - repeated Embedding batch_embeddings = 6; + repeated Embedding batch_embeddings = 5; } message Embedding {