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

restore progress_callback behavior

Browse source
This commit is contained in:
slaren 2023-12-20 23:15:15 +01:00
parent 5834a25345
commit ecb23d4ac5
1 changed files with 5 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

9
llama.cpp
View file

@ -2372,6 +2372,11 @@ struct llama_model_loader {
for (int i = 0; i < gguf_get_n_tensors(ctx_gguf); i++) {
struct ggml_tensor * cur = ggml_get_tensor(ctx, gguf_get_tensor_name(ctx_gguf, i));
GGML_ASSERT(cur); // unused tensors should have been caught by load_data already
if (progress_callback) {
progress_callback((float) size_done / size_data, progress_callback_user_data);
}
const size_t offs = file_offset(ggml_get_name(cur));
if (!legacy_offload || cur->backend == GGML_BACKEND_CPU) {
@ -2422,10 +2427,6 @@ struct llama_model_loader {
}
size_done += ggml_nbytes(cur);
if (progress_callback) {
progress_callback((float) size_done / size_data, progress_callback_user_data);
}
}
// unmap GPU tensors