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

clean up a bit

Browse source
This commit is contained in:
ngxson 2024-06-12 16:08:27 +02:00
parent a2a5f1bfbd
commit b22c8459ff
3 changed files with 11 additions and 27 deletions
Download patch file Download diff file Expand all files Collapse all files

14
common/common.cpp
View file

@ -1982,13 +1982,13 @@ void gpt_params_print_usage(int /*argc*/, char ** argv, const gpt_params & param
#endif // LOG_DISABLE_LOGS
options.push_back({ "control-vector-generator" });
options.push_back({ "control-vector-generator", "-o, --output FNAME", "output file (default: '%s')", params.cvector_outfile.c_str() });
options.push_back({ "control-vector-generator", "--positive-file FNAME", "positive prompts file, one prompt per line (default: '%s')", params.cvector_positive_file.c_str() });
options.push_back({ "control-vector-generator", "--negative-file FNAME", "negative prompts file, one prompt per line (default: '%s')", params.cvector_negative_file.c_str() });
options.push_back({ "control-vector-generator", "--completions-file", "completions file (default: '%s')", params.cvector_completions_file.c_str() });
options.push_back({ "control-vector-generator", "--num-completions N", "number of lines of completions file to use (default: %d)", params.n_completions });
options.push_back({ "control-vector-generator", "--batch-pca N", "batch size used for PCA. Larger batch runs faster, but uses more memory (default: %d)", params.n_pca_batch });
options.push_back({ "control-vector-generator", "--iter-pca N", "number of iterations used for PCA (default: %d)", params.n_pca_iterations });
options.push_back({ "control-vector-generator", "-o, --output FNAME", "output file (default: '%s')", params.cvector_outfile.c_str() });
options.push_back({ "control-vector-generator", "--positive-file FNAME", "positive prompts file, one prompt per line (default: '%s')", params.cvector_positive_file.c_str() });
options.push_back({ "control-vector-generator", "--negative-file FNAME", "negative prompts file, one prompt per line (default: '%s')", params.cvector_negative_file.c_str() });
options.push_back({ "control-vector-generator", "--completions-file FNAME","completions file (default: '%s')", params.cvector_completions_file.c_str() });
options.push_back({ "control-vector-generator", "--num-completions N", "number of lines of completions file to use (default: %d)", params.n_completions });
options.push_back({ "control-vector-generator", "--batch-pca N", "batch size used for PCA. Larger batch runs faster, but uses more memory (default: %d)", params.n_pca_batch });
options.push_back({ "control-vector-generator", "--iter-pca N", "number of iterations used for PCA (default: %d)", params.n_pca_iterations });
printf("usage: %s [options]\n", argv[0]);

14
examples/control-vector-generator/control-vector-generator.cpp
View file

@ -213,6 +213,7 @@ struct train_context {
}
// build the v_diff tensors from v_diff_tmp (v_diff need to be transposed)
// TODO @ngxson : maybe add option NOT to transpose v_diff; will be useful for "mean" method
void build_v_diff() {
printf("build_v_diff\n");
for (int il = 0; il < n_layers - 1; il++) {
@ -228,7 +229,6 @@ struct train_context {
for (int ir = 0; ir < n_rows; ++ir) {
for (int ic = 0; ic < n_embd; ++ic) {
float f = arr[ir*n_embd + ic];
//std::cout << ir << "," << ic << " = " << f << "\n";
ggml_set_f32_nd(diff, ir, ic, 0, 0, f);
}
}
@ -341,12 +341,9 @@ static void export_gguf(const std::vector<struct ggml_tensor *> & v_ctrl, const
printf("Added tensor: %s\n", v_ctrl[i]->name);
}
printf("Writing file...\n");
printf("%s: writing file...\n", __func__);
gguf_write_to_file(ctx, fname.c_str(), false);
printf("%s: wrote file '%s'\n", __func__, fname.c_str());
gguf_free(ctx);
}
@ -370,11 +367,8 @@ static int prepare_entries(gpt_params & params, train_context & ctx_train) {
// create templated prompts
std::vector<std::string> completions = ctrlvec_load_prompt_file(params.cvector_completions_file, false);
auto format_template = [](std::string persona, std::string suffix) {
//const std::string user_tag = "[INST]";
//const std::string asst_tag = "[/INST]";
//return user_tag + " Act as if you're extremely " + persona + ". " + asst_tag + " " + suffix;
// TODO make this dynamic - allow the user to change it somehow - and adapt based on model
return persona + " " + suffix; // entry in positive/negative.txt must already be formatted i.e. "[INST] Act as if you're extremely happy. [/INST]"
// entry in positive/negative.txt must already be formatted i.e. "[INST] Act as if you're extremely happy. [/INST]"
return persona + " " + suffix;
};
for (size_t i = 0; i < positive_prompts.size(); ++i) {
for (int j = 0; j < std::min((int) completions.size(), params.n_completions); ++j) {

10
examples/control-vector-generator/pca.hpp
View file

@ -123,15 +123,6 @@ struct pca_model {
}
ggml_backend_tensor_set(dev_eigenvector, random_vec.data(), 0, ggml_nbytes(dev_eigenvector));
}
// init host context
//struct ggml_init_params host_params = {
// /*.mem_size =*/ (n_embd * sizeof(float) + ggml_tensor_overhead()) * 2u,
// /*.mem_buffer =*/ NULL,
// /*.no_alloc =*/ false,
//};
//ctx_host = ggml_init(host_params);
//host_eigenvector = ggml_new_tensor_1d(ctx_host, GGML_TYPE_F32, n_embd);
}
~pca_model() {
@ -322,7 +313,6 @@ static void run_pca(
params.n_layers = v_input.size();
power_iteration(params, v_input[il], ctrl_out);
printf("%s: Done layer %d / %d\n", __func__, (int) il+1, (int) v_input.size());
//print_debug_tensor(ctrl_out);
}
}