starred/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-04-26 02:36:09 +02:00
Code Issues Packages Projects Releases Wiki Activity

ggml update to b7108 (#12992)

Browse Source 
* Revert "vulkan: temporary cary of vulkan fixes (#12971)"

This reverts commit 3a9e8e9fd4.

* ggml update to b7087

* fix argsort on metal

* update to b7108

* fix bakllava regression

This model lacks the metadata for the projector type.

* update to b7209

* fix TopK perf

* only build arm code on arm
This commit is contained in:
Daniel Hiltgen
2025-12-03 19:43:29 -08:00
committed by GitHub
parent 854d40edc5
commit 0cf7794b16
303 changed files with 32711 additions and 23435 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
llama/llama.cpp/src/llama-grammar.cpp vendored
View File

@@ -6,8 +6,10 @@
#include <cmath>
#include <algorithm>
#include <cstdint>
#include <stdexcept>
#define MAX_REPETITION_THRESHOLD 2000
//
// helpers
//
@@ -345,8 +347,10 @@ const char * llama_grammar_parser::parse_sequence(
size_t last_sym_start = rule.size();
const char * pos = src;
auto handle_repetitions = [&](int min_times, int max_times) {
// use UINT64_MAX as the empty value because we aligned to the proper uint64_t type so -1 can't be used
// (though it's technically the same as -1 now)
auto handle_repetitions = [&](uint64_t min_times, uint64_t max_times) {
bool no_max = max_times == UINT64_MAX;
if (last_sym_start == rule.size()) {
throw std::runtime_error(std::string("expecting preceding item to */+/?/{ at ") + pos);
}
@@ -373,20 +377,20 @@ const char * llama_grammar_parser::parse_sequence(
rule.resize(last_sym_start);
} else {
// Repeat the previous elements (min_times - 1) times
for (int i = 1; i < min_times; i++) {
for (uint64_t i = 1; i < min_times; i++) {
rule.insert(rule.end(), prev_rule.begin(), prev_rule.end());
}
}
uint32_t last_rec_rule_id = 0;
auto n_opt = max_times < 0 ? 1 : max_times - min_times;
auto n_opt = no_max ? 1 : max_times - min_times;
llama_grammar_rule rec_rule(prev_rule);
for (int i = 0; i < n_opt; i++) {
for (uint64_t i = 0; i < n_opt; i++) {
rec_rule.resize(prev_rule.size());
uint32_t rec_rule_id = generate_symbol_id( rule_name);
if (i > 0 || max_times < 0) {
rec_rule.push_back({LLAMA_GRETYPE_RULE_REF, max_times < 0 ? rec_rule_id : last_rec_rule_id});
if (i > 0 || no_max) {
rec_rule.push_back({LLAMA_GRETYPE_RULE_REF, no_max ? rec_rule_id : last_rec_rule_id});
}
rec_rule.push_back({LLAMA_GRETYPE_ALT, 0});
rec_rule.push_back({LLAMA_GRETYPE_END, 0});
@@ -478,10 +482,10 @@ const char * llama_grammar_parser::parse_sequence(
throw std::runtime_error(std::string("expecting an int at ") + pos);
}
const char * int_end = parse_int(pos);
int min_times = std::stoul(std::string(pos, int_end - pos));
uint64_t min_times = std::stoul(std::string(pos, int_end - pos));
pos = parse_space(int_end, is_nested);
int max_times = -1;
uint64_t max_times = UINT64_MAX; // default: no max limit
if (*pos == '}') {
max_times = min_times;
@@ -502,6 +506,10 @@ const char * llama_grammar_parser::parse_sequence(
} else {
throw std::runtime_error(std::string("expecting ',' at ") + pos);
}
bool has_max = max_times != UINT64_MAX;
if (min_times > MAX_REPETITION_THRESHOLD || (has_max && max_times > MAX_REPETITION_THRESHOLD)) {
throw std::runtime_error(std::string("number of repetitions exceeds sane defaults, please reduce the number of repetitions"));
}
handle_repetitions(min_times, max_times);
} else {
break;