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

llama : remove cfg smooth factor as it is only a reparameterization of the guidance scale (#2280)

Browse source
This commit is contained in:
Guillaume "Vermeille" Sanchez 2023-07-21 12:58:36 +02:00 • committed by GitHub
parent 73643f5fb1
commit ab0e26bdfb
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
5 changed files with 4 additions and 24 deletions
Download patch file Download diff file Expand all files Collapse all files

7
examples/common.cpp
View file

@ -260,12 +260,6 @@ bool gpt_params_parse(int argc, char ** argv, gpt_params & params) {
break;
}
params.cfg_scale = std::stof(argv[i]);
} else if (arg == "--cfg-smooth-factor") {
if (++i >= argc) {
invalid_param = true;
break;
}
params.cfg_smooth_factor = std::stof(argv[i]);
} else if (arg == "-b" || arg == "--batch-size") {
if (++i >= argc) {
invalid_param = true;
@ -509,7 +503,6 @@ void gpt_print_usage(int /*argc*/, char ** argv, const gpt_params & params) {
fprintf(stderr, " --cfg-negative-prompt PROMPT \n");
fprintf(stderr, " negative prompt to use for guidance. (default: empty)\n");
fprintf(stderr, " --cfg-scale N strength of guidance (default: %f, 1.0 = disable)\n", params.cfg_scale);
fprintf(stderr, " --cfg-smooth-factor N smooth factor between old and new logits (default: %f, 1.0 = no smoothing)\n", params.cfg_smooth_factor);
fprintf(stderr, " -c N, --ctx-size N size of the prompt context (default: %d)\n", params.n_ctx);
fprintf(stderr, " --rope-freq-base N RoPE base frequency (default: %.1f)\n", params.rope_freq_base);
fprintf(stderr, " --rope-freq-scale N RoPE frequency scaling factor (default: %g)\n", params.rope_freq_scale);

1
examples/common.h
View file

@ -55,7 +55,6 @@ struct gpt_params {
// https://arxiv.org/abs/2306.17806
std::string cfg_negative_prompt; // string to help guidance
float cfg_scale = 1.f; // How strong is guidance
float cfg_smooth_factor = 1.f; // Smooth factor between old and new logits
std::string model = "models/7B/ggml-model.bin"; // model path
std::string model_alias = "unknown"; // model alias

2
examples/main/main.cpp
View file

@ -557,7 +557,7 @@ int main(int argc, char ** argv) {
llama_token_data_array candidates_p = { candidates.data(), candidates.size(), false };
if (ctx_guidance) {
llama_sample_classifier_free_guidance(ctx, &candidates_p, ctx_guidance, params.cfg_scale, params.cfg_smooth_factor);
llama_sample_classifier_free_guidance(ctx, &candidates_p, ctx_guidance, params.cfg_scale);
}
// Apply penalties