vrr/prima.cpp
2
0
Fork 0
mirror of https://github.com/Lizonghang/prima.cpp.git synced 2025-09-09 14:34:34 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

use highs to solve the allocation program

Browse source
This commit is contained in:
Lizonghang 2025-01-15 10:04:04 +04:00
parent b577c10d25
commit 5d9aadf3d5
6 changed files with 614 additions and 86 deletions
Download patch file Download diff file Expand all files Collapse all files

27
include/llama.h
View file

@ -442,12 +442,12 @@ extern "C" {
LLAMA_API void llama_free_model(struct llama_model * model);
LLAMA_API void llama_init_sockets (struct llama_context * ctx, uint32_t n_world, uint32_t my_rank);
LLAMA_API void llama_free_sockets (struct llama_context * ctx, char ** msg);
LLAMA_API int llama_gather_device_info (struct llama_context * ctx, struct device_info * dev_info_set);
LLAMA_API int llama_send_device_info (struct llama_context * ctx, struct device_info * dev_info);
LLAMA_API int llama_broadcast_n_layer_window(struct llama_context * ctx, uint32_t * n_layer_window);
LLAMA_API int llama_recv_n_layer_window(struct llama_context * ctx, uint32_t * n_layer_window);
LLAMA_API void llama_init_sockets (struct llama_context * ctx, uint32_t n_world, uint32_t my_rank);
LLAMA_API void llama_free_sockets (struct llama_context * ctx, char ** msg);
LLAMA_API int llama_gather_device_info(struct llama_context * ctx, struct device_info * dev_info_set);
LLAMA_API int llama_send_device_info (struct llama_context * ctx, struct device_info * dev_info);
LLAMA_API int llama_bcast_layer_setup (struct llama_context * ctx, uint32_t * n_layer_window, uint32_t * n_gpu_layers);
LLAMA_API int llama_recv_layer_setup (struct llama_context * ctx, uint32_t * n_layer_window, uint32_t * n_gpu_layers);
LLAMA_API int llm_load_tensors(
struct llama_model_loader * ml,
@ -465,6 +465,8 @@ extern "C" {
LLAMA_API uint32_t * llama_context_n_layer_window(struct llama_context * ctx);
LLAMA_API uint32_t * llama_context_n_gpu_layers(struct llama_context * ctx);
// Frees all allocated memory
LLAMA_API void llama_free(struct llama_context * ctx);
@ -536,11 +538,14 @@ extern "C" {
// Return the size of compute buffer size, including input tensors and activations
LLAMA_API void llama_model_compute_buf_size(
uint64_t * cpu_buf,
uint64_t * gpu_buf,
int64_t * cpu_buf,
int64_t * gpu_buf,
const struct llama_model * model,
const struct llama_context_params cparams,
bool use_gpu);
bool use_gpu,
bool is_master,
int n_layers,
int n_gpu_layers);
// Return the size of KV cache in the model
LLAMA_API void llama_total_kv_size(
@ -551,8 +556,8 @@ extern "C" {
bool use_gpu);
LLAMA_API void llama_kv_size(
uint64_t * cpu_cache,
uint64_t * gpu_cache,
int64_t * cpu_cache,
int64_t * gpu_cache,
const struct llama_model * model,
const struct llama_context_params cparams,
bool use_gpu);