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

Record sampling time in llama_sample_classifier_free_guidance

Browse source
This commit is contained in:
Bach Le 2023-07-08 00:07:49 +08:00
parent 66eb048470
commit 8e66e59cdd
1 changed files with 6 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

6
llama.cpp
View file

@ -2163,6 +2163,8 @@ void llama_sample_classifier_free_guidance(
struct llama_context * guidance_ctx,
float scale,
float smooth_factor) {
int64_t t_start_sample_us = t_start_sample_us = ggml_time_us();
assert(ctx);
auto n_vocab = llama_n_vocab(ctx);
assert(n_vocab == (int)candidates->size);
@ -2195,6 +2197,10 @@ void llama_sample_classifier_free_guidance(
candidates->data[i].logit = smooth_factor * guidance_logit + (1.f - smooth_factor) * base_logit;
}
if (ctx) {
ctx->t_sample_us += ggml_time_us() - t_start_sample_us;
}
}
llama_token llama_sample_token_mirostat(struct llama_context * ctx, llama_token_data_array * candidates, float tau, float eta, int m, float * mu) {