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

llama : fix save/load state context size

Browse source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2023-11-02 19:02:35 +02:00
parent 83c96d5809
commit 8401e3ebcd
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
1 changed files with 2 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
llama.cpp
View file

@ -8558,7 +8558,7 @@ static void llama_copy_state_data_internal(struct llama_context * ctx, llama_dat
if (kv_buf_size) {
const size_t elt_size = ggml_element_size(kv_self.k);
ggml_context * cpy_ctx = ggml_init({ 4096, NULL, /* no_alloc */ true });
ggml_context * cpy_ctx = ggml_init({ 6*ggml_tensor_overhead() + ggml_graph_overhead(), NULL, /* no_alloc */ true });
ggml_cgraph * gf = ggml_new_graph(cpy_ctx);
ggml_tensor * kout3d = ggml_new_tensor_3d(cpy_ctx, kv_self.k->type, n_embd, kv_head, n_layer);
@ -8686,7 +8686,7 @@ size_t llama_set_state_data(struct llama_context * ctx, uint8_t * src) {
const size_t elt_size = ggml_element_size(kv_self.k);
ggml_context * cpy_ctx = ggml_init({ 4096, NULL, /* no_alloc */ true });
ggml_context * cpy_ctx = ggml_init({ 6*ggml_tensor_overhead() + ggml_graph_overhead(), NULL, /* no_alloc */ true });
ggml_cgraph * gf = ggml_new_graph(cpy_ctx);
ggml_tensor * kin3d = ggml_new_tensor_3d(cpy_ctx, kv_self.k->type, n_embd, kv_head, n_layer);