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

llama: initialize f16 tables in quantize c api.

Browse source
This commit is contained in:
Yaohui Liu 2023-05-20 00:32:08 +08:00
parent 5ea4339273
commit 484f6e9438
No known key found for this signature in database
GPG key ID: E86D01E1809BD23E
2 changed files with 6 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

7
examples/quantize/quantize.cpp
View file

@ -52,13 +52,6 @@ int main(int argc, char ** argv) {
return 1; return 1;
} }
// needed to initialize f16 tables
{
struct ggml_init_params params = { 0, NULL, false };
struct ggml_context * ctx = ggml_init(params);
ggml_free(ctx);
}
// parse command line arguments // parse command line arguments
const std::string fname_inp = argv[1]; const std::string fname_inp = argv[1];
std::string fname_out; std::string fname_out;

6
llama.cpp
View file

@ -2198,6 +2198,12 @@ int llama_model_quantize(
enum llama_ftype ftype, enum llama_ftype ftype,
int nthread) { int nthread) {
try { try {
// needed to initialize f16 tables
{
struct ggml_init_params params = { 0, NULL, false };
struct ggml_context * ctx = ggml_init(params);
ggml_free(ctx);
}
llama_model_quantize_internal(fname_inp, fname_out, ftype, nthread); llama_model_quantize_internal(fname_inp, fname_out, ftype, nthread);
return 0; return 0;
} catch (const std::string & err) { } catch (const std::string & err) {