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

fix for typical sampler

Browse source
This commit is contained in:
Concedo 2023-09-01 15:24:00 +08:00
parent 0c3a265187
commit d7fed4732f
2 changed files with 9 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

6
llama.cpp
View file

@ -1234,6 +1234,7 @@ struct llama_model_loader {
// determine file type based on the number of tensors for each quantization and print meta data
// TODO: make optional
if(false) //disable this log for now
{
std::map<enum ggml_type, uint32_t> n_type;
@ -3974,7 +3975,10 @@ void llama_sample_typical(struct llama_context * ctx, llama_token_data_array * c
float entropy = 0.0f;
for (size_t i = 0; i < candidates->size; ++i) {
entropy += -candidates->data[i].p * logf(candidates->data[i].p);
if(candidates->data[i].p>0)
{
entropy += -candidates->data[i].p * logf(candidates->data[i].p);
}
}
// Compute the absolute difference between negative log probability and entropy for each candidate

5
otherarch/llama_v3.cpp
View file

@ -2663,7 +2663,10 @@ void llama_v3_sample_typical(struct llama_v3_context * ctx, llama_v3_token_data_
float entropy = 0.0f;
for (size_t i = 0; i < candidates->size; ++i) {
entropy += -candidates->data[i].p * logf(candidates->data[i].p);
if(candidates->data[i].p>0)
{
entropy += -candidates->data[i].p * logf(candidates->data[i].p);
}
}
// Compute the absolute difference between negative log probability and entropy for each candidate