root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-25 02:44:36 +00:00
Code Issues Actions 9 Packages Projects Releases Wiki Activity

ggml : force no_alloc == false when creating opt tensors (close #1699)

Browse Source 
This is needed to make operators like ggml_view() be able to store their
parameters in the ggml context's memory and not get discarded when
no_alloc is true
This commit is contained in:
Georgi Gerganov 2023-06-10 12:06:45 +03:00
parent e9b66ee982
commit 17c10acfb4
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
1 changed files with 9 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
ggml.c
View File

@ -3721,6 +3721,7 @@ struct ggml_context {
void * mem_buffer;
bool mem_buffer_owned;
bool no_alloc;
bool no_alloc_save; // this is used to save the no_alloc state when using scratch buffers
int n_objects;
@ -4055,6 +4056,7 @@ struct ggml_context * ggml_init(struct ggml_init_params params) {
/*.mem_buffer =*/ params.mem_buffer ? params.mem_buffer : GGML_ALIGNED_MALLOC(mem_size),
/*.mem_buffer_owned =*/ params.mem_buffer ? false : true,
/*.no_alloc =*/ params.no_alloc,
/*.no_alloc_save =*/ params.no_alloc,
/*.n_objects =*/ 0,
/*.objects_begin =*/ NULL,
/*.objects_end =*/ NULL,
@ -4132,11 +4134,18 @@ size_t ggml_get_mem_size(struct ggml_context * ctx) {
// operators when using scratch buffers
// TODO: implement a better way
void ggml_scratch_save(struct ggml_context * ctx) {
// this is needed to allow opt tensors to store their data
// TODO: again, need to find a better way
ctx->no_alloc_save = ctx->no_alloc;
ctx->no_alloc = false;
ctx->scratch_save = ctx->scratch;
ctx->scratch.data = NULL;
}
void ggml_scratch_load(struct ggml_context * ctx) {
ctx->no_alloc = ctx->no_alloc_save;
ctx->scratch = ctx->scratch_save;
}