llama : support NUL bytes in tokens

This commit is contained in:
Francis Couture-Harpin 2024-08-11 21:00:03 -04:00
parent 4134999e01
commit faaac59d16
7 changed files with 28 additions and 18 deletions

View file

@ -561,7 +561,7 @@ static void load_vocab(const char * filename, const Config * config, struct llam
vocab->id_to_token.resize(n_vocab);
for (uint32_t i = 0; i < n_vocab; i++) {
std::string word = gguf_get_arr_str(ctx, token_idx, i);
std::string word(gguf_get_arr_str(ctx, token_idx, i), gguf_get_arr_str_n(ctx, token_idx, i));
vocab->token_to_id[word] = i;

View file

@ -12,7 +12,7 @@ static bool g_verbose = false;
static std::string get_kv_str(struct gguf_context * ctx_gguf, const std::string & key){
int id = gguf_find_key(ctx_gguf, key.c_str());
return id < 0 ? "" : std::string(gguf_get_val_str(ctx_gguf, id));
return id < 0 ? "" : std::string(gguf_get_val_str(ctx_gguf, id), gguf_get_val_str_n(ctx_gguf, id));
}
static float get_kv_f32(struct gguf_context * ctx_gguf, const std::string & key) {

View file

@ -225,7 +225,7 @@ static std::string gguf_kv_to_str(const struct gguf_context * ctx_gguf, int i) {
switch (type) {
case GGUF_TYPE_STRING:
return gguf_get_val_str(ctx_gguf, i);
return std::string(gguf_get_val_str(ctx_gguf, i), gguf_get_val_str_n(ctx_gguf, i));
case GGUF_TYPE_ARRAY:
{
const enum gguf_type arr_type = gguf_get_arr_type(ctx_gguf, i);
@ -235,7 +235,7 @@ static std::string gguf_kv_to_str(const struct gguf_context * ctx_gguf, int i) {
ss << "[";
for (int j = 0; j < arr_n; j++) {
if (arr_type == GGUF_TYPE_STRING) {
std::string val = gguf_get_arr_str(ctx_gguf, i, j);
std::string val(gguf_get_arr_str(ctx_gguf, i, j), gguf_get_arr_str_n(ctx_gguf, i, j));
// escape quotes
replace_all(val, "\\", "\\\\");
replace_all(val, "\"", "\\\"");