002f9879b2
Replaces cherry-picked upstream_ref with complete source trees. xllm/ — Iluvatar official C++ inference engine (15MB, 1470 files) Complete: kernels → layers → models → runtime → scheduler → api Excluded: .git, binary images, third_party submodule checkouts ds_vllm/ — Iluvatar official vllm fork (8MB, 703 files) Included: csrc/ (ALL CUDA kernels), fused_moe/, qwen3_5 model, _custom_ops Excluded: tests, benchmarks, docs, examples (not needed for reference) Critical call chains now fully traceable: MoE: moe_topk_softmax_kernels.cuh → ixformer.h → fused_moe.cpp → layer GDN: qwen3_gated_delta_net_base.cpp → qwen3_5_gated_delta_net.cpp Attention: ixformer.h → xllm_paged_attention → attention.cpp
90 lines
1.6 KiB
Protocol Buffer
90 lines
1.6 KiB
Protocol Buffer
syntax = "proto3";
|
|
|
|
option go_package = "jd.com/jd-infer/xllm;xllm";
|
|
package xllm.proto;
|
|
|
|
import "google/protobuf/struct.proto";
|
|
|
|
message Function {
|
|
string name = 1;
|
|
string description = 2;
|
|
google.protobuf.Struct parameters = 3;
|
|
}
|
|
|
|
message Tool {
|
|
string type = 1; // "function"
|
|
Function function = 2;
|
|
}
|
|
|
|
message ToolCall {
|
|
optional uint32 index = 1;
|
|
optional string id = 2;
|
|
string type = 3; // "function"
|
|
FunctionCall function = 4;
|
|
}
|
|
|
|
message FunctionCall {
|
|
string name = 1;
|
|
string arguments = 2; // JSON string
|
|
}
|
|
|
|
enum Priority {
|
|
DEFAULT = 0;
|
|
|
|
HIGH = 1;
|
|
|
|
NORMAL = 2;
|
|
|
|
LOW = 3;
|
|
}
|
|
|
|
|
|
message Usage {
|
|
// the number of tokens in the prompt.
|
|
optional int32 prompt_tokens = 1 [json_name="prompt_tokens"];
|
|
|
|
// the number of tokens in the generated completion.
|
|
optional int32 completion_tokens = 2 [json_name="completion_tokens"];
|
|
|
|
// the total number of tokens used in the request (prompt + completion).
|
|
optional int32 total_tokens = 3 [json_name="total_tokens"];
|
|
}
|
|
|
|
// Options for streaming response.
|
|
message StreamOptions{
|
|
// if set, an additional chunk with usage will be streamed before the data: [DONE] message.
|
|
optional bool include_usage = 1;
|
|
}
|
|
|
|
message Status {
|
|
bool ok = 1;
|
|
}
|
|
|
|
message Routing {
|
|
string prefill_name = 1;
|
|
|
|
string decode_name = 2;
|
|
}
|
|
|
|
message Empty {}
|
|
|
|
message AddressInfo {
|
|
// the ip adderss of worker process.
|
|
string address = 1;
|
|
int32 global_rank = 2;
|
|
}
|
|
|
|
message CommUniqueId {
|
|
bytes comm_unique_id = 1;
|
|
}
|
|
|
|
message CommUniqueIdList {
|
|
repeated CommUniqueId comm_unique_ids = 1;
|
|
}
|
|
|
|
enum MasterStatus {
|
|
WAKEUP = 0;
|
|
LIGHT_SLEEP = 1;
|
|
DEEP_SLEEP = 2;
|
|
}
|