vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

fixed memory leak by freeing temporary graph during session load

Browse source
This commit is contained in:
l3utterfly 2023-10-08 00:36:20 +08:00
parent 88a14fcfef
commit 9ee8aeccd7
1 changed files with 4 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

4
llama.cpp
View file

@ -8405,6 +8405,10 @@ size_t llama_set_state_data(struct llama_context * ctx, uint8_t * src) {
ggml_graph_compute_helper(ctx->work_buffer, gf, /*n_threads*/ 1); ggml_graph_compute_helper(ctx->work_buffer, gf, /*n_threads*/ 1);
ggml_free(cpy_ctx); ggml_free(cpy_ctx);
// free our allocated graph
free(gf);
gf = NULL;
} }
ctx->kv_self.head = kv_head; ctx->kv_self.head = kv_head;