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

llama : make tensor_split ptr instead of array (#2272)

Browse source
This commit is contained in:
Georgi Gerganov 2023-07-21 13:10:51 +03:00 • committed by GitHub
parent 54e3bc76fe
commit ae178ab46b
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
4 changed files with 8 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

4
llama.cpp
View file

@ -849,7 +849,7 @@ struct llama_context_params llama_context_default_params() {
/*.n_batch =*/ 512,
/*.gpu_layers =*/ 0,
/*.main_gpu =*/ 0,
/*.tensor_split =*/ {0},
/*.tensor_split =*/ nullptr,
/*.rope_freq_base =*/ 10000.0f,
/*.rope_freq_scale =*/ 1.0f,
/*.progress_callback =*/ nullptr,
@ -1289,7 +1289,7 @@ static bool llama_model_load(
int n_batch,
int n_gpu_layers,
int main_gpu,
float * tensor_split,
const float * tensor_split,
float rope_freq_base,
float rope_freq_scale,
bool low_vram,