grammar: increase MAX_REPETITION_THRESHOLD + make it configurable via envvar

src/llama-grammar.cpp

@@ -7,10 +7,22 @@
 #include <cmath>
 #include <algorithm>
 #include <cstdint>
+#include <cstdlib>
 #include <set>
 #include <stdexcept>
 
-#define MAX_REPETITION_THRESHOLD 2000
+static constexpr uint64_t DEFAULT_MAX_REPETITION_THRESHOLD = 50000;
+
+llama_grammar_parser::llama_grammar_parser(const struct llama_vocab * vocab)
+    : vocab(vocab) {
+    const char * env = std::getenv("LLAMA_GRAMMAR_MAX_REPS");
+    if (env) {
+        max_repetition_threshold = std::stoull(env);
+    } else {
+        max_repetition_threshold = DEFAULT_MAX_REPETITION_THRESHOLD;
+    }
+}
+
 //
 // helpers
 //
@@ -491,7 +503,7 @@ const char * llama_grammar_parser::parse_sequence(
             total_rules = min_times;
         }
 
-        if (n_prev_rules * total_rules >= MAX_REPETITION_THRESHOLD) {
+        if (n_prev_rules * total_rules >= max_repetition_threshold) {
             throw std::runtime_error("number of rules that are going to be repeated multiplied by the new repetition exceeds sane defaults, please reduce the number of repetitions or rule complexity");
         }
 
@@ -649,7 +661,7 @@ const char * llama_grammar_parser::parse_sequence(
                 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)) {
+            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);

src/llama-grammar.h

@@ -89,7 +89,10 @@ struct llama_grammar_parser {
 
     llama_grammar_rules rules;
 
-    llama_grammar_parser(const struct llama_vocab * vocab = nullptr) : vocab(vocab) {}
+    uint64_t max_repetition_threshold;
+
+    llama_grammar_parser(const struct llama_vocab * vocab = nullptr);
+
 
     llama_grammar_stack c_rules() const;
 

tests/test-grammar-parser.cpp

@@ -146,7 +146,7 @@ int main()
     )""");
 
     verify_failure(R"""(
-        root ::= (((((([^x]*){0,99}){0,99}){0,99}){0,99}){0,99}){0,99}
+        root ::= (((((([^x]*){0,999}){0,999}){0,999}){0,999}){0,999}){0,999}
     )""");
 
     verify_failure(R"""(