matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-14 09:47:02 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama: remove model loading for grammar (#10096)

Browse Source
This commit is contained in:
Parth Sareen
2025-04-24 11:51:19 -07:00
committed by GitHub
parent 40b10eee6d
commit a53d744b01
13 changed files with 521 additions and 107 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
llama/llama.cpp/src/llama-grammar.h vendored
View File

@@ -6,8 +6,19 @@
#include <regex>
#include <string>
#include <vector>
#include <set>
struct llama_vocab;
struct ollama_vocab {
std::map<uint32_t, std::string> token_to_piece_map;
std::set<uint32_t> special_eog_ids;
const std::string & token_to_piece(const uint32_t token) const;
void add_token_pieces(const uint32_t* tokens, size_t n_tokens, const char** pieces);
void set_eog_tokens(const uint32_t* tokens, size_t n_tokens);
bool is_eog(const uint32_t token) const;
};
// grammar element type
enum llama_gretype {
@@ -114,6 +125,7 @@ struct llama_grammar_trigger_pattern {
struct llama_grammar {
// note: allow null vocab for testing (not great)
const llama_vocab * vocab;
const ollama_vocab * o_vocab;
const llama_grammar_rules rules; // TODO: shared ptr
llama_grammar_stacks stacks;
@@ -141,12 +153,14 @@ struct llama_grammar {
// note: needed for tests (not great)
struct llama_grammar * llama_grammar_init_impl(
const struct llama_vocab * vocab,
const struct ollama_vocab * ollama_vocab,
const llama_grammar_element ** rules,
size_t n_rules,
size_t start_rule_index);
struct llama_grammar * llama_grammar_init_impl(
const struct llama_vocab * vocab,
const struct ollama_vocab * ollama_vocab,
const char * grammar_str,
const char * grammar_root,
bool lazy,