char-fork/llama.cpp

git clone https://git.t4t.associates/char-fork/llama.cpp

Radoslav Gerganovrpc : implement event and async backend APIs (#18626)d0132a680

char/aliceai-t5-moe
brancheschar/aliceai-t5-moeall refs →
1.2 KiB35 linesraw
1#pragma once
2
3#include "ggml-backend.h"
4
5#ifdef  __cplusplus
6extern "C" {
7#endif
8
9#define RPC_PROTO_MAJOR_VERSION    6
10#define RPC_PROTO_MINOR_VERSION    0
11#define RPC_PROTO_PATCH_VERSION    0
12
13#ifdef  __cplusplus
14static_assert(GGML_OP_COUNT == 101, "GGML_OP_COUNT has changed - update RPC_PROTO_PATCH_VERSION");
15#endif
16
17#define GGML_RPC_MAX_SERVERS       16
18
19// backend API
20GGML_BACKEND_API ggml_backend_t ggml_backend_rpc_init(const char * endpoint, uint32_t device);
21GGML_BACKEND_API bool ggml_backend_is_rpc(ggml_backend_t backend);
22
23GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_rpc_buffer_type(const char * endpoint, uint32_t device);
24
25GGML_BACKEND_API void ggml_backend_rpc_get_device_memory(const char * endpoint, uint32_t device, size_t * free, size_t * total);
26
27GGML_BACKEND_API void ggml_backend_rpc_start_server(const char * endpoint, const char * cache_dir,
28                                                    size_t n_threads, size_t n_devices, ggml_backend_dev_t * devices);
29
30GGML_BACKEND_API ggml_backend_reg_t ggml_backend_rpc_reg(void);
31GGML_BACKEND_API ggml_backend_reg_t ggml_backend_rpc_add_server(const char * endpoint);
32
33#ifdef  __cplusplus
34}
35#endif