github/gpt4all
1
0
Fork 0
mirror of https://github.com/nomic-ai/gpt4all.git synced 2025-09-05 18:40:50 +00:00
Code Issues Packages Projects Releases Wiki Activity

backend: update to latest commit of llama.cpp Vulkan PR

Browse Source 
Signed-off-by: Jared Van Bortel <jared@nomic.ai>
This commit is contained in:
Jared Van Bortel
2024-01-25 16:58:46 -05:00
committed by AT
parent 29d2c936d1
commit 38c61493d2
9 changed files with 85 additions and 125 deletions
Download Patch File Download Diff File Expand all files Collapse all files

45
gpt4all-backend/llmodel_shared.h
View File

@@ -4,50 +4,6 @@
#include <vector>
#include <ggml.h>
#if defined(GGML_USE_KOMPUTE)
#include "ggml-kompute.h"
struct llm_buffer {
uint8_t * addr = NULL;
size_t size = 0;
ggml_vk_memory memory;
bool force_cpu = false;
llm_buffer() = default;
void resize(size_t size) {
free();
if (!ggml_vk_has_device() || force_cpu) {
this->addr = new uint8_t[size];
this->size = size;
} else {
this->memory = ggml_vk_allocate(size);
this->addr = (uint8_t*)memory.data;
this->size = size;
}
}
void free() {
if (!memory.primaryMemory) {
delete[] addr;
} else if (memory.data) {
ggml_vk_free_memory(memory);
}
this->addr = NULL;
this->size = 0;
}
~llm_buffer() {
free();
}
// disable copy and move
llm_buffer(const llm_buffer&) = delete;
llm_buffer(llm_buffer&&) = delete;
llm_buffer& operator=(const llm_buffer&) = delete;
llm_buffer& operator=(llm_buffer&&) = delete;
};
#else
struct llm_buffer {
uint8_t * addr = NULL;
size_t size = 0;
@@ -62,7 +18,6 @@ struct llm_buffer {
delete[] addr;
}
};
#endif
struct llm_kv_cache {
struct ggml_tensor * k;