sglang 0.5.3rc2__py3-none-any.whl → 0.5.4__py3-none-any.whl

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\"\xe1\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\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\nconstraintB\x11\n\x0f_max_new_tokens\"]\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\"\x14\n\x12HealthCheckRequest\"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)
@@ -39,73 +39,81 @@ if not _descriptor._USE_C_DESCRIPTORS:
   _globals['_SAMPLINGPARAMS_LOGITBIASENTRY']._loaded_options = None
   _globals['_SAMPLINGPARAMS_LOGITBIASENTRY']._serialized_options = b'8\001'
   _globals['_SAMPLINGPARAMS']._serialized_start=113
-  _globals['_SAMPLINGPARAMS']._serialized_end=850
-  _globals['_SAMPLINGPARAMS_LOGITBIASENTRY']._serialized_start=769
-  _globals['_SAMPLINGPARAMS_LOGITBIASENTRY']._serialized_end=817
-  _globals['_DISAGGREGATEDPARAMS']._serialized_start=852
-  _globals['_DISAGGREGATEDPARAMS']._serialized_end=945
-  _globals['_GENERATEREQUEST']._serialized_start=948
-  _globals['_GENERATEREQUEST']._serialized_end=1558
-  _globals['_TOKENIZEDINPUT']._serialized_start=1560
-  _globals['_TOKENIZEDINPUT']._serialized_end=1618
-  _globals['_MULTIMODALINPUTS']._serialized_start=1621
-  _globals['_MULTIMODALINPUTS']._serialized_end=1832
-  _globals['_GENERATERESPONSE']._serialized_start=1835
-  _globals['_GENERATERESPONSE']._serialized_end=2062
-  _globals['_GENERATESTREAMCHUNK']._serialized_start=2065
-  _globals['_GENERATESTREAMCHUNK']._serialized_end=2342
-  _globals['_GENERATECOMPLETE']._serialized_start=2345
-  _globals['_GENERATECOMPLETE']._serialized_end=2756
-  _globals['_GENERATEERROR']._serialized_start=2758
-  _globals['_GENERATEERROR']._serialized_end=2833
-  _globals['_OUTPUTLOGPROBS']._serialized_start=2835
-  _globals['_OUTPUTLOGPROBS']._serialized_end=2952
-  _globals['_INPUTLOGPROBS']._serialized_start=2955
-  _globals['_INPUTLOGPROBS']._serialized_end=3113
-  _globals['_INPUTTOKENLOGPROB']._serialized_start=3115
-  _globals['_INPUTTOKENLOGPROB']._serialized_end=3164
-  _globals['_TOPLOGPROBS']._serialized_start=3166
-  _globals['_TOPLOGPROBS']._serialized_end=3214
-  _globals['_HIDDENSTATES']._serialized_start=3216
-  _globals['_HIDDENSTATES']._serialized_end=3279
-  _globals['_EMBEDREQUEST']._serialized_start=3282
-  _globals['_EMBEDREQUEST']._serialized_end=3612
-  _globals['_EMBEDRESPONSE']._serialized_start=3615
-  _globals['_EMBEDRESPONSE']._serialized_end=3772
-  _globals['_EMBEDCOMPLETE']._serialized_start=3775
-  _globals['_EMBEDCOMPLETE']._serialized_end=3938
-  _globals['_EMBEDDING']._serialized_start=3940
-  _globals['_EMBEDDING']._serialized_end=3982
-  _globals['_EMBEDERROR']._serialized_start=3984
-  _globals['_EMBEDERROR']._serialized_end=4044
-  _globals['_HEALTHCHECKREQUEST']._serialized_start=4046
-  _globals['_HEALTHCHECKREQUEST']._serialized_end=4124
-  _globals['_HEALTHCHECKRESPONSE']._serialized_start=4126
-  _globals['_HEALTHCHECKRESPONSE']._serialized_end=4181
-  _globals['_ABORTREQUEST']._serialized_start=4183
-  _globals['_ABORTREQUEST']._serialized_end=4233
-  _globals['_ABORTRESPONSE']._serialized_start=4235
-  _globals['_ABORTRESPONSE']._serialized_end=4284
-  _globals['_LOADLORAREQUEST']._serialized_start=4286
-  _globals['_LOADLORAREQUEST']._serialized_end=4359
-  _globals['_LOADLORARESPONSE']._serialized_start=4361
-  _globals['_LOADLORARESPONSE']._serialized_end=4433
-  _globals['_UNLOADLORAREQUEST']._serialized_start=4435
-  _globals['_UNLOADLORAREQUEST']._serialized_end=4474
-  _globals['_UNLOADLORARESPONSE']._serialized_start=4476
-  _globals['_UNLOADLORARESPONSE']._serialized_end=4530
-  _globals['_UPDATEWEIGHTSREQUEST']._serialized_start=4532
-  _globals['_UPDATEWEIGHTSREQUEST']._serialized_end=4651
-  _globals['_UPDATEWEIGHTSRESPONSE']._serialized_start=4653
-  _globals['_UPDATEWEIGHTSRESPONSE']._serialized_end=4710
-  _globals['_GETINTERNALSTATEREQUEST']._serialized_start=4712
-  _globals['_GETINTERNALSTATEREQUEST']._serialized_end=4757
-  _globals['_GETINTERNALSTATERESPONSE']._serialized_start=4759
-  _globals['_GETINTERNALSTATERESPONSE']._serialized_end=4825
-  _globals['_SETINTERNALSTATEREQUEST']._serialized_start=4827
-  _globals['_SETINTERNALSTATEREQUEST']._serialized_end=4892
-  _globals['_SETINTERNALSTATERESPONSE']._serialized_start=4894
-  _globals['_SETINTERNALSTATERESPONSE']._serialized_end=4954
-  _globals['_SGLANGSCHEDULER']._serialized_start=4957
-  _globals['_SGLANGSCHEDULER']._serialized_end=5339
+  _globals['_SAMPLINGPARAMS']._serialized_end=833
+  _globals['_SAMPLINGPARAMS_LOGITBIASENTRY']._serialized_start=732
+  _globals['_SAMPLINGPARAMS_LOGITBIASENTRY']._serialized_end=780
+  _globals['_DISAGGREGATEDPARAMS']._serialized_start=835
+  _globals['_DISAGGREGATEDPARAMS']._serialized_end=928
+  _globals['_GENERATEREQUEST']._serialized_start=931
+  _globals['_GENERATEREQUEST']._serialized_end=1541
+  _globals['_TOKENIZEDINPUT']._serialized_start=1543
+  _globals['_TOKENIZEDINPUT']._serialized_end=1601
+  _globals['_MULTIMODALINPUTS']._serialized_start=1604
+  _globals['_MULTIMODALINPUTS']._serialized_end=1815
+  _globals['_GENERATERESPONSE']._serialized_start=1818
+  _globals['_GENERATERESPONSE']._serialized_end=2045
+  _globals['_GENERATESTREAMCHUNK']._serialized_start=2048
+  _globals['_GENERATESTREAMCHUNK']._serialized_end=2325
+  _globals['_GENERATECOMPLETE']._serialized_start=2328
+  _globals['_GENERATECOMPLETE']._serialized_end=2739
+  _globals['_GENERATEERROR']._serialized_start=2741
+  _globals['_GENERATEERROR']._serialized_end=2816
+  _globals['_OUTPUTLOGPROBS']._serialized_start=2818
+  _globals['_OUTPUTLOGPROBS']._serialized_end=2935
+  _globals['_INPUTLOGPROBS']._serialized_start=2938
+  _globals['_INPUTLOGPROBS']._serialized_end=3096
+  _globals['_INPUTTOKENLOGPROB']._serialized_start=3098
+  _globals['_INPUTTOKENLOGPROB']._serialized_end=3147
+  _globals['_TOPLOGPROBS']._serialized_start=3149
+  _globals['_TOPLOGPROBS']._serialized_end=3197
+  _globals['_HIDDENSTATES']._serialized_start=3199
+  _globals['_HIDDENSTATES']._serialized_end=3262
+  _globals['_EMBEDREQUEST']._serialized_start=3265
+  _globals['_EMBEDREQUEST']._serialized_end=3595
+  _globals['_EMBEDRESPONSE']._serialized_start=3598
+  _globals['_EMBEDRESPONSE']._serialized_end=3755
+  _globals['_EMBEDCOMPLETE']._serialized_start=3758
+  _globals['_EMBEDCOMPLETE']._serialized_end=3921
+  _globals['_EMBEDDING']._serialized_start=3923
+  _globals['_EMBEDDING']._serialized_end=3965
+  _globals['_EMBEDERROR']._serialized_start=3967
+  _globals['_EMBEDERROR']._serialized_end=4027
+  _globals['_HEALTHCHECKREQUEST']._serialized_start=4029
+  _globals['_HEALTHCHECKREQUEST']._serialized_end=4049
+  _globals['_HEALTHCHECKRESPONSE']._serialized_start=4051
+  _globals['_HEALTHCHECKRESPONSE']._serialized_end=4106
+  _globals['_ABORTREQUEST']._serialized_start=4108
+  _globals['_ABORTREQUEST']._serialized_end=4158
+  _globals['_ABORTRESPONSE']._serialized_start=4160
+  _globals['_ABORTRESPONSE']._serialized_end=4209
+  _globals['_LOADLORAREQUEST']._serialized_start=4211
+  _globals['_LOADLORAREQUEST']._serialized_end=4284
+  _globals['_LOADLORARESPONSE']._serialized_start=4286
+  _globals['_LOADLORARESPONSE']._serialized_end=4358
+  _globals['_UNLOADLORAREQUEST']._serialized_start=4360
+  _globals['_UNLOADLORAREQUEST']._serialized_end=4399
+  _globals['_UNLOADLORARESPONSE']._serialized_start=4401
+  _globals['_UNLOADLORARESPONSE']._serialized_end=4455
+  _globals['_UPDATEWEIGHTSREQUEST']._serialized_start=4457
+  _globals['_UPDATEWEIGHTSREQUEST']._serialized_end=4576
+  _globals['_UPDATEWEIGHTSRESPONSE']._serialized_start=4578
+  _globals['_UPDATEWEIGHTSRESPONSE']._serialized_end=4635
+  _globals['_GETINTERNALSTATEREQUEST']._serialized_start=4637
+  _globals['_GETINTERNALSTATEREQUEST']._serialized_end=4682
+  _globals['_GETINTERNALSTATERESPONSE']._serialized_start=4684
+  _globals['_GETINTERNALSTATERESPONSE']._serialized_end=4750
+  _globals['_SETINTERNALSTATEREQUEST']._serialized_start=4752
+  _globals['_SETINTERNALSTATEREQUEST']._serialized_end=4817
+  _globals['_SETINTERNALSTATERESPONSE']._serialized_start=4819
+  _globals['_SETINTERNALSTATERESPONSE']._serialized_end=4879
+  _globals['_GETMODELINFOREQUEST']._serialized_start=4881
+  _globals['_GETMODELINFOREQUEST']._serialized_end=4902
+  _globals['_GETMODELINFORESPONSE']._serialized_start=4905
+  _globals['_GETMODELINFORESPONSE']._serialized_end=5267
+  _globals['_GETSERVERINFOREQUEST']._serialized_start=5269
+  _globals['_GETSERVERINFOREQUEST']._serialized_end=5291
+  _globals['_GETSERVERINFORESPONSE']._serialized_start=5294
+  _globals['_GETSERVERINFORESPONSE']._serialized_end=5605
+  _globals['_SGLANGSCHEDULER']._serialized_start=5608
+  _globals['_SGLANGSCHEDULER']._serialized_end=6203
 # @@protoc_insertion_point(module_scope)

sglang/srt/grpc/sglang_scheduler_pb2.pyi

@@ -11,7 +11,7 @@ from typing import ClassVar as _ClassVar, Optional as _Optional, Union as _Union
 DESCRIPTOR: _descriptor.FileDescriptor
 
 class SamplingParams(_message.Message):
-    __slots__ = ("temperature", "top_p", "top_k", "min_p", "frequency_penalty", "presence_penalty", "repetition_penalty", "max_new_tokens", "stop", "stop_token_ids", "skip_special_tokens", "spaces_between_special_tokens", "regex", "json_schema", "ebnf_grammar", "structural_tag", "lora_path", "n", "token_healing", "min_new_tokens", "ignore_eos", "no_stop_trim", "stream_interval", "logit_bias", "custom_params")
+    __slots__ = ("temperature", "top_p", "top_k", "min_p", "frequency_penalty", "presence_penalty", "repetition_penalty", "max_new_tokens", "stop", "stop_token_ids", "skip_special_tokens", "spaces_between_special_tokens", "regex", "json_schema", "ebnf_grammar", "structural_tag", "n", "min_new_tokens", "ignore_eos", "no_stop_trim", "stream_interval", "logit_bias", "custom_params")
     class LogitBiasEntry(_message.Message):
         __slots__ = ("key", "value")
         KEY_FIELD_NUMBER: _ClassVar[int]
@@ -35,9 +35,7 @@ class SamplingParams(_message.Message):
     JSON_SCHEMA_FIELD_NUMBER: _ClassVar[int]
     EBNF_GRAMMAR_FIELD_NUMBER: _ClassVar[int]
     STRUCTURAL_TAG_FIELD_NUMBER: _ClassVar[int]
-    LORA_PATH_FIELD_NUMBER: _ClassVar[int]
     N_FIELD_NUMBER: _ClassVar[int]
-    TOKEN_HEALING_FIELD_NUMBER: _ClassVar[int]
     MIN_NEW_TOKENS_FIELD_NUMBER: _ClassVar[int]
     IGNORE_EOS_FIELD_NUMBER: _ClassVar[int]
     NO_STOP_TRIM_FIELD_NUMBER: _ClassVar[int]
@@ -60,16 +58,14 @@ class SamplingParams(_message.Message):
     json_schema: str
     ebnf_grammar: str
     structural_tag: str
-    lora_path: str
     n: int
-    token_healing: bool
     min_new_tokens: int
     ignore_eos: bool
     no_stop_trim: bool
     stream_interval: int
     logit_bias: _containers.ScalarMap[str, float]
     custom_params: _struct_pb2.Struct
-    def __init__(self, temperature: _Optional[float] = ..., top_p: _Optional[float] = ..., top_k: _Optional[int] = ..., min_p: _Optional[float] = ..., frequency_penalty: _Optional[float] = ..., presence_penalty: _Optional[float] = ..., repetition_penalty: _Optional[float] = ..., max_new_tokens: _Optional[int] = ..., stop: _Optional[_Iterable[str]] = ..., stop_token_ids: _Optional[_Iterable[int]] = ..., skip_special_tokens: bool = ..., spaces_between_special_tokens: bool = ..., regex: _Optional[str] = ..., json_schema: _Optional[str] = ..., ebnf_grammar: _Optional[str] = ..., structural_tag: _Optional[str] = ..., lora_path: _Optional[str] = ..., n: _Optional[int] = ..., token_healing: bool = ..., min_new_tokens: _Optional[int] = ..., ignore_eos: bool = ..., no_stop_trim: bool = ..., stream_interval: _Optional[int] = ..., logit_bias: _Optional[_Mapping[str, float]] = ..., custom_params: _Optional[_Union[_struct_pb2.Struct, _Mapping]] = ...) -> None: ...
+    def __init__(self, temperature: _Optional[float] = ..., top_p: _Optional[float] = ..., top_k: _Optional[int] = ..., min_p: _Optional[float] = ..., frequency_penalty: _Optional[float] = ..., presence_penalty: _Optional[float] = ..., repetition_penalty: _Optional[float] = ..., max_new_tokens: _Optional[int] = ..., stop: _Optional[_Iterable[str]] = ..., stop_token_ids: _Optional[_Iterable[int]] = ..., skip_special_tokens: bool = ..., spaces_between_special_tokens: bool = ..., regex: _Optional[str] = ..., json_schema: _Optional[str] = ..., ebnf_grammar: _Optional[str] = ..., structural_tag: _Optional[str] = ..., n: _Optional[int] = ..., min_new_tokens: _Optional[int] = ..., ignore_eos: bool = ..., no_stop_trim: bool = ..., stream_interval: _Optional[int] = ..., logit_bias: _Optional[_Mapping[str, float]] = ..., custom_params: _Optional[_Union[_struct_pb2.Struct, _Mapping]] = ...) -> None: ...
 
 class DisaggregatedParams(_message.Message):
     __slots__ = ("bootstrap_host", "bootstrap_port", "bootstrap_room")
@@ -324,10 +320,8 @@ class EmbedError(_message.Message):
     def __init__(self, message: _Optional[str] = ..., code: _Optional[str] = ..., details: _Optional[str] = ...) -> None: ...
 
 class HealthCheckRequest(_message.Message):
-    __slots__ = ("tokenized",)
-    TOKENIZED_FIELD_NUMBER: _ClassVar[int]
-    tokenized: TokenizedInput
-    def __init__(self, tokenized: _Optional[_Union[TokenizedInput, _Mapping]] = ...) -> None: ...
+    __slots__ = ()
+    def __init__(self) -> None: ...
 
 class HealthCheckResponse(_message.Message):
     __slots__ = ("healthy", "message")
@@ -432,3 +426,65 @@ 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: ...

sglang/srt/grpc/sglang_scheduler_pb2_grpc.py

@@ -8,7 +8,7 @@ import warnings
 
 from . import sglang_scheduler_pb2 as sglang__scheduler__pb2
 
-GRPC_GENERATED_VERSION = '1.74.0'
+GRPC_GENERATED_VERSION = '1.75.1'
 GRPC_VERSION = grpc.__version__
 _version_not_supported = False
 
@@ -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)

sglang/srt/layers/activation.py

@@ -380,4 +380,7 @@ if not (
     logger.info(
         "sgl-kernel is not available on Non-NV, Non-AMD platforms or Non-AMX CPUs. Fallback to other kernel libraries."
     )
-    from vllm.model_executor.layers.activation import GeluAndMul, SiluAndMul
+    from vllm.model_executor.layers.activation import (  # noqa: F401
+        GeluAndMul,
+        SiluAndMul,
+    )

sglang/srt/layers/attention/aiter_backend.py

@@ -1064,7 +1064,7 @@ class AiterMultiStepDraftBackend:
             device=model_runner.device,
         )
         self.attn_backends = []
-        for i in range(self.speculative_num_steps):
+        for i in range(self.speculative_num_steps - 1):
             self.attn_backends.append(
                 AiterAttnBackend(
                     model_runner,
@@ -1107,7 +1107,7 @@ class AiterMultiStepDraftBackend:
             self.page_size,
         )
 
-        for i in range(self.speculative_num_steps):
+        for i in range(self.speculative_num_steps - 1):
             forward_batch.spec_info.kv_indptr = self.kv_indptr[i, : bs + 1]
             forward_batch.spec_info.kv_indices = kv_indices_buffer[i][
                 : seq_lens_sum * self.topk + bs * (i + 1)
@@ -1141,7 +1141,7 @@ class AiterMultiStepDraftBackend:
             dtype=torch.int32,
             device=self.device,
         )
-        for i in range(self.speculative_num_steps):
+        for i in range(self.speculative_num_steps - 1):
             self.attn_backends[i].init_cuda_graph_state(
                 max_bs, max_num_tokens, kv_indices_buf=self.cuda_graph_kv_indices[i]
             )

sglang/srt/layers/attention/ascend_backend.py

@@ -20,7 +20,6 @@ if TYPE_CHECKING:
     from sglang.srt.layers.radix_attention import RadixAttention
     from sglang.srt.model_executor.model_runner import ModelRunner
 
-import os
 
 import numpy as np
 
@@ -356,6 +355,11 @@ class AscendAttnBackend(AttentionBackend):
             assert (
                 layer.qk_head_dim != layer.v_head_dim
             ), "FIA only supports qk_head_dim != v_head_dim"
+            num_token_padding = q.shape[0]
+            q, k, v = [
+                data[: forward_batch.num_token_non_padded_cpu] for data in [q, k, v]
+            ]
+
             q_nope, q_rope = q.split([layer.v_head_dim, self.qk_rope_head_dim], dim=-1)
             k_nope, k_rope = k.split([layer.v_head_dim, self.qk_rope_head_dim], dim=-1)
 
@@ -375,6 +379,18 @@ class AscendAttnBackend(AttentionBackend):
                 next_tokens=0,
             )
 
+            attn_output = attn_output.reshape(-1, layer.tp_q_head_num, layer.v_head_dim)
+            if num_token_padding != forward_batch.num_token_non_padded_cpu:
+                attn_output = torch.cat(
+                    [
+                        attn_output,
+                        attn_output.new_zeros(
+                            num_token_padding - attn_output.shape[0],
+                            *attn_output.shape[1:],
+                        ),
+                    ],
+                    dim=0,
+                )
         return attn_output
 
     def forward_decode_graph(

sglang/srt/layers/attention/attention_registry.py

@@ -1,7 +1,14 @@
 import logging
+from typing import TYPE_CHECKING
 
 logger = logging.getLogger(__name__)
 
+
+if TYPE_CHECKING:
+    # evade circular imports
+    from sglang.srt.layers.attention.base_attn_backend import AttentionBackend
+    from sglang.srt.model_executor.model_runner import ModelRunner
+
 ATTENTION_BACKENDS = {}
 
 
@@ -27,7 +34,9 @@ def create_flashinfer_backend(runner):
                 or not runner.plan_stream_for_flashinfer
             ):
                 runner.plan_stream_for_flashinfer = torch.cuda.Stream()
-        return FlashInferAttnBackend(runner)
+        return FlashInferAttnBackend(
+            runner, init_new_workspace=runner.init_new_workspace
+        )
     else:
         from sglang.srt.layers.attention.flashinfer_mla_backend import (
             FlashInferMLAAttnBackend,
@@ -129,9 +138,6 @@ def create_flashattention_v3_backend(runner):
 
 @register_attention_backend("fa4")
 def create_flashattention_v4_backend(runner):
-    assert (
-        runner.use_mla_backend
-    ), "FlashAttention v4 Support is at an early stage, only MLA model supported now"
     from sglang.srt.layers.attention.flashattention_backend import FlashAttentionBackend
 
     return FlashAttentionBackend(runner, fa_impl_ver=4)
@@ -169,38 +175,52 @@ def create_dual_chunk_flash_attn_backend(runner):
     return DualChunkFlashAttentionBackend(runner)
 
 
-def attn_backend_wrapper(runner, full_attn_backend):
+def attn_backend_wrapper(runner: "ModelRunner", full_attn_backend: "AttentionBackend"):
     """
     Wrapper for special models like hybrid GDN, so we don't
     need to change the code of the original attention backend.
     """
     assert not (
-        runner.is_hybrid_gdn and runner.use_mla_backend
+        runner.hybrid_gdn_config is not None and runner.use_mla_backend
     ), "hybrid_gdn can only be used with non-MLA models."
 
-    # wrap for hybrid GDN models
-    if runner.is_hybrid_gdn:
-        from sglang.srt.utils import is_blackwell, is_npu
-
-        if is_blackwell():
-            assert (
-                runner.server_args.attention_backend == "triton"
-                or runner.server_args.attention_backend == "trtllm_mha"
-            ), "triton or trtllm_mha backend are the only supported backends on Blackwell GPUs for hybrid GDN models, use --attention-backend triton or --attention-backend trtllm_mha to specify the backend."
-        if is_npu():
-            assert (
-                runner.server_args.attention_backend == "ascend"
-            ), "ascend backend is the only supported backend on NPU for hybrid GDN models, use --attention-backend ascend to specify the backend."
-        logger.info(f"Using hybrid linear attention backend for hybrid GDN models.")
+    if cfg := runner.mambaish_config:
+        from sglang.srt.layers.attention.fla.utils import check_environments
         from sglang.srt.layers.attention.hybrid_linear_attn_backend import (
+            GDNAttnBackend,
             HybridLinearAttnBackend,
-            MambaAttnBackend,
+            Mamba2AttnBackend,
         )
+        from sglang.srt.utils import is_blackwell, is_npu
 
-        linear_attn_backend = MambaAttnBackend(runner)
-        full_attn_layers = runner.model_config.hf_config.full_attention_layer_ids
+        check_environments()
+        if runner.hybrid_gdn_config is not None:
+            if is_blackwell():
+                assert (
+                    runner.server_args.attention_backend == "triton"
+                ), "triton backend is the only supported backend on Blackwell GPUs for hybrid GDN models, use --attention-backend triton to specify the backend."
+            if is_npu():
+                assert (
+                    runner.server_args.attention_backend == "ascend"
+                ), "ascend backend is the only supported backend on NPU for hybrid GDN models, use --attention-backend ascend to specify the backend."
+            logger.info(f"Using hybrid linear attention backend for hybrid GDN models.")
+            linear_attn_backend = GDNAttnBackend(runner)
+        elif runner.mamba2_config is not None:
+            linear_attn_backend = Mamba2AttnBackend(runner)
+        else:
+            raise ValueError(
+                "Expected hybrid GDN or NemotronH models, but got unknown model."
+            )
+        full_attn_layers = cfg.full_attention_layer_ids
         return HybridLinearAttnBackend(
             full_attn_backend, linear_attn_backend, full_attn_layers
         )
 
     return full_attn_backend
+
+
+@register_attention_backend("intel_xpu")
+def create_intel_xpu_backend(runner):
+    from sglang.srt.layers.attention.xpu_backend import XPUAttentionBackend
+
+    return XPUAttentionBackend(runner)