llama.cpp/ggml-mpi.h

40 lines
911 B
C
Raw Normal View History

#pragma once
struct ggml_context;
struct ggml_tensor;
struct ggml_cgraph;
#ifdef __cplusplus
extern "C" {
#endif
struct ggml_mpi_context;
void ggml_mpi_backend_init(void);
void ggml_mpi_backend_free(void);
struct ggml_mpi_context * ggml_mpi_init(void);
void ggml_mpi_free(struct ggml_mpi_context * ctx);
int ggml_mpi_rank(struct ggml_mpi_context * ctx);
void ggml_mpi_eval_init(
struct ggml_mpi_context * ctx_mpi,
int * n_tokens,
int * n_past,
int * n_threads);
void ggml_mpi_graph_compute_pre(
struct ggml_mpi_context * ctx_mpi,
struct ggml_cgraph * gf,
int n_layers);
void ggml_mpi_graph_compute_post(
struct ggml_mpi_context * ctx_mpi,
struct ggml_cgraph * gf,
int n_layers);
#ifdef __cplusplus
}
#endif