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

Simplify vector building logic

Browse source
This commit is contained in:
ochafik 2023-08-23 22:21:07 +01:00
parent 5132130af7
commit 463e117820
1 changed files with 9 additions and 13 deletions
Download patch file Download diff file Expand all files Collapse all files

22
examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp
View file

@ -623,20 +623,16 @@ void save_as_llama_model(struct llama_vocab * vocab, struct my_llama_model * mod
struct gguf_context * ctx = gguf_init_empty(); struct gguf_context * ctx = gguf_init_empty();
std::vector<const char*> tokens; std::vector<const char*> tokens;
std::transform(vocab->id_to_token.begin(), vocab->id_to_token.end(), std::back_inserter(tokens),
[](const llama_vocab::token_data & token_data) { return token_data.text.c_str(); });
gguf_set_arr_str(ctx, "tokenizer.ggml.tokens", tokens.data(), tokens.size());
std::vector<float> scores; std::vector<float> scores;
std::transform(vocab->id_to_token.begin(), vocab->id_to_token.end(), std::back_inserter(scores), std::vector<llama_token_type> token_types;
[](const llama_vocab::token_data & token_data) { return token_data.score; }); for (const llama_vocab::token_data & token_data : vocab->id_to_token) {
gguf_set_arr_data(ctx, "tokenizer.ggml.scores", GGUF_TYPE_FLOAT32, scores.data(), scores.size()); tokens.push_back(token_data.text.c_str());
scores.push_back(token_data.score);
std::vector<int> token_types; token_types.push_back(token_data.type);
for (size_t i = 0; i < vocab->id_to_token.size(); ++i) {
// TODO: Refine this.
token_types.push_back(LLAMA_TOKEN_TYPE_UNDEFINED);
} }
// TODO(ochafik): Do we need to output merges too, maybe?
gguf_set_arr_str(ctx, "tokenizer.ggml.tokens", tokens.data(), tokens.size());
gguf_set_arr_data(ctx, "tokenizer.ggml.scores", GGUF_TYPE_FLOAT32, scores.data(), scores.size());
gguf_set_arr_data(ctx, "tokenizer.ggml.token_type", GGUF_TYPE_INT32, token_types.data(), token_types.size()); gguf_set_arr_data(ctx, "tokenizer.ggml.token_type", GGUF_TYPE_INT32, token_types.data(), token_types.size());
gguf_set_val_str(ctx, "tokenizer.ggml.model", "llama"); gguf_set_val_str(ctx, "tokenizer.ggml.model", "llama");
@ -669,7 +665,7 @@ void save_as_llama_model(struct llama_vocab * vocab, struct my_llama_model * mod
gguf_add_tensor(ctx, model->norm); gguf_add_tensor(ctx, model->norm);
ggml_set_name(model->output, TN_OUTPUT); ggml_set_name(model->output, TN_OUTPUT);
gguf_add_tensor(ctx, model->output); // ? gguf_add_tensor(ctx, model->output);
for (uint32_t i = 0; i < model->hparams.n_layer; ++i) { for (uint32_t i = 0; i < model->hparams.n_layer; ++i) {
auto & layer = model->layers[i]; auto & layer = model->layers[i];