thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 4
llama-cpp-turboquant/tests/test-llama-grammar.cpp
Andrea Arcangeli 990e4d9698
common/grammar: fix grammar parsing issues to prevent stack overflow and hangs (#18604) 
* grammar: add test case for nullable symbol loop

Reproduce stack overflow (or OOM) with ( [x]* )* found while adding
GBNF support to ripgrep-edit.

llama-server reproducer:

curl \
  -X POST \
  -d '{
    "messages": [{ "role": "user", "content": "write yes" }],
    "grammar": "root ::= ( [x]* )*"
  }' \
  -H "Content-Type: application/json" \
  http://localhost:8811/v1/chat/completions

* grammar: prevent stack overflow with nullable symbol loop

Fix a potential stack overflow in llama_grammar_advance_stack that
could occur when processing grammars with nullable symbols that lead
to infinite derivations of empty strings. The fix introduces cycle
detection by tracking visited stacks to prevent infinite recursion.

rg-edit regexp: llama_grammar_advance_stack
rg-edit extra-args: -A20
rg-edit directive: """Rewrite: fix the following segfault:

[..]
⚫ Testing segfault. Grammar:
            root ::= ( [x]* )*

            root ::= ( [x]* )*

Segmentation fault         build/bin/test-grammar-integration"""

gptel-context:
(("~/llama.cpp/src/llama-grammar.cpp")
 ("~/llama.cpp/tests/test-grammar-integration.cpp")
 ("~/llama.cpp/grammars/./list.gbnf")
 ("~/llama.cpp/grammars/./json_arr.gbnf")
 ("~/llama.cpp/grammars/./json.gbnf")
 ("~/llama.cpp/grammars/./japanese.gbnf")
 ("~/llama.cpp/grammars/./english.gbnf")
 ("~/llama.cpp/grammars/./chess.gbnf")
 ("~/llama.cpp/grammars/./c.gbnf")
 ("~/llama.cpp/grammars/./arithmetic.gbnf")
 ("~/llama.cpp/grammars/./README.md"))

* grammar: convert recursive llama_grammar_advance_stack to iterative

This change converts the function to an iterative approach using
explicit stacks, which prevents deep recursion and eliminates the risk
of stack overflow.

rg-edit regexp: llama_grammar_advance_stack
rg-edit extra-args: -A30
rg-edit directive: """Rewrite: fix the following segfault:

[..]
⚫ Testing segfault. Grammar:
            root ::= ( [x]* )*

            root ::= ( [x]* )*

Segmentation fault         build/bin/test-grammar-integration

convert from recursive to interactive"""

gptel-context:
(("~/llama.cpp/src/llama-grammar.cpp")
 ("~/llama.cpp/tests/test-grammar-integration.cpp")
 ("~/llama.cpp/grammars/./list.gbnf")
 ("~/llama.cpp/grammars/./json_arr.gbnf")
 ("~/llama.cpp/grammars/./json.gbnf")
 ("~/llama.cpp/grammars/./japanese.gbnf")
 ("~/llama.cpp/grammars/./english.gbnf")
 ("~/llama.cpp/grammars/./chess.gbnf")
 ("~/llama.cpp/grammars/./c.gbnf")
 ("~/llama.cpp/grammars/./arithmetic.gbnf")
 ("~/llama.cpp/grammars/./README.md"))

v2: Added a `std::set` to perform tree-based lookups with O(N log N)
complexity. Testing with a parallel run of `test-grammar-integration`
shows a double-digit percentage increase in runtime. An
`unordered_set` with O(1) hashing was also evaluated, but the overhead
of constructing hash keys from pointers made it significantly slower
than the rbtree implementation that only requires an ordering
operator. The performance regression in the test suite appears
justified by the overall reduction in algorithmic complexity.

Co-developed-by: Piotr Wilkin (ilintar) <piotr.wilkin@syndatis.com>

* grammar: add test case for hang in repetition grammar processing

This commit adds a new test case to the grammar integration tests that
specifically targets a hang scenario in the repetition grammar parser
found while adding GBNF support to ripgrep-edit.

llama-server reproducer:

curl \
  -X POST \
  -d '{
    "messages": [{ "role": "user", "content": "write yes" }],
    "grammar": "root ::= (([^x]*){0,99}){0,99}"
  }' \
  -H "Content-Type: application/json" \
  http://localhost:8811/v1/chat/completions

* grammar: add repetition threshold check

The change introduces a maximum repetition threshold to avoid
excessive rule expansion during grammar parsing. When parsing
repetition patterns like {m,n}, the parser now calculates the
potential number of rules that would be generated and throws an error
if the product of previous rules and new rules exceeds the threshold.

A test case was added to verify the threshold is properly enforced for
deeply nested repetition patterns that would otherwise cause hangs.
2026-03-21 18:43:35 +01:00

406 lines
11 KiB
C++
Raw Blame History

#ifdef NDEBUG
#undef NDEBUG
#endif
#include "llama.h"
#include "../src/llama-grammar.h"
#include <cassert>
#include <stdexcept>
int main()
{
llama_grammar_parser parsed_grammar;
std::vector<std::pair<std::string, uint32_t>> expected = {
{"expr", 2},
{"expr_6", 6},
{"expr_7", 7},
{"ident", 8},
{"ident_10", 10},
{"num", 9},
{"num_11", 11},
{"root", 0},
{"root_1", 1},
{"root_5", 5},
{"term", 4},
{"ws", 3},
{"ws_12", 12},
};
std::vector<std::vector<llama_grammar_element>> expected_rules = {
{{LLAMA_GRETYPE_RULE_REF, 5}, {LLAMA_GRETYPE_END, 0}},
{
{LLAMA_GRETYPE_RULE_REF, 2},
{LLAMA_GRETYPE_CHAR, 61},
{LLAMA_GRETYPE_RULE_REF, 3},
{LLAMA_GRETYPE_RULE_REF, 4},
{LLAMA_GRETYPE_CHAR, 10},
{LLAMA_GRETYPE_END, 0},
},
{{LLAMA_GRETYPE_RULE_REF, 4}, {LLAMA_GRETYPE_RULE_REF, 7}, {LLAMA_GRETYPE_END, 0}},
{{LLAMA_GRETYPE_RULE_REF, 12}, {LLAMA_GRETYPE_END, 0}},
{
{LLAMA_GRETYPE_RULE_REF, 8},
{LLAMA_GRETYPE_ALT, 0},
{LLAMA_GRETYPE_RULE_REF, 9},
{LLAMA_GRETYPE_ALT, 0},
{LLAMA_GRETYPE_CHAR, 40},
{LLAMA_GRETYPE_RULE_REF, 3},
{LLAMA_GRETYPE_RULE_REF, 2},
{LLAMA_GRETYPE_CHAR, 41},
{LLAMA_GRETYPE_RULE_REF, 3},
{LLAMA_GRETYPE_END, 0},
},
{{LLAMA_GRETYPE_RULE_REF, 1}, {LLAMA_GRETYPE_RULE_REF, 5}, {LLAMA_GRETYPE_ALT, 0}, {LLAMA_GRETYPE_RULE_REF, 1}, {LLAMA_GRETYPE_END, 0}},
{
{LLAMA_GRETYPE_CHAR, 45},
{LLAMA_GRETYPE_CHAR_ALT, 43},
{LLAMA_GRETYPE_CHAR_ALT, 42},
{LLAMA_GRETYPE_CHAR_ALT, 47},
{LLAMA_GRETYPE_RULE_REF, 4},
{LLAMA_GRETYPE_END, 0},
},
{{LLAMA_GRETYPE_RULE_REF, 6}, {LLAMA_GRETYPE_RULE_REF, 7}, {LLAMA_GRETYPE_ALT, 0}, {LLAMA_GRETYPE_END, 0}},
{
{LLAMA_GRETYPE_CHAR, 97},
{LLAMA_GRETYPE_CHAR_RNG_UPPER, 122},
{LLAMA_GRETYPE_RULE_REF, 10},
{LLAMA_GRETYPE_RULE_REF, 3},
{LLAMA_GRETYPE_END, 0},
},
{{LLAMA_GRETYPE_RULE_REF, 11}, {LLAMA_GRETYPE_RULE_REF, 3}, {LLAMA_GRETYPE_END, 0}},
{
{LLAMA_GRETYPE_CHAR, 97},
{LLAMA_GRETYPE_CHAR_RNG_UPPER, 122},
{LLAMA_GRETYPE_CHAR_ALT, 48},
{LLAMA_GRETYPE_CHAR_RNG_UPPER, 57},
{LLAMA_GRETYPE_CHAR_ALT, 95},
{LLAMA_GRETYPE_RULE_REF, 10},
{LLAMA_GRETYPE_ALT, 0},
{LLAMA_GRETYPE_END, 0},
},
{
{LLAMA_GRETYPE_CHAR, 48},
{LLAMA_GRETYPE_CHAR_RNG_UPPER, 57},
{LLAMA_GRETYPE_RULE_REF, 11},
{LLAMA_GRETYPE_ALT, 0},
{LLAMA_GRETYPE_CHAR, 48},
{LLAMA_GRETYPE_CHAR_RNG_UPPER, 57},
{LLAMA_GRETYPE_END, 0},
},
{
{LLAMA_GRETYPE_CHAR, 32},
{LLAMA_GRETYPE_CHAR_ALT, 9},
{LLAMA_GRETYPE_CHAR_ALT, 10},
{LLAMA_GRETYPE_RULE_REF, 12},
{LLAMA_GRETYPE_ALT, 0},
{LLAMA_GRETYPE_END, 0},
},
};
for (auto pair : expected)
{
parsed_grammar.symbol_ids[pair.first] = pair.second;
}
for (auto rule : expected_rules)
{
parsed_grammar.rules.emplace_back();
for (auto element : rule)
{
parsed_grammar.rules.back().push_back(element);
}
}
std::vector<const llama_grammar_element *> grammar_rules(parsed_grammar.c_rules());
llama_grammar * grammar = llama_grammar_init_impl(nullptr, grammar_rules.data(), grammar_rules.size(), parsed_grammar.symbol_ids.at("root"));
if (grammar == nullptr) {
throw std::runtime_error("Failed to initialize llama_grammar");
}
std::vector<std::vector<llama_grammar_element>> expected_stacks = {
{
{LLAMA_GRETYPE_CHAR, 61},
{LLAMA_GRETYPE_RULE_REF, 7},
{LLAMA_GRETYPE_CHAR, 40},
},
{
{LLAMA_GRETYPE_CHAR, 61},
{LLAMA_GRETYPE_RULE_REF, 7},
{LLAMA_GRETYPE_RULE_REF, 3},
{LLAMA_GRETYPE_CHAR, 48},
},
{
{LLAMA_GRETYPE_CHAR, 61},
{LLAMA_GRETYPE_RULE_REF, 7},
{LLAMA_GRETYPE_RULE_REF, 3},
{LLAMA_GRETYPE_CHAR, 48},
},
{
{LLAMA_GRETYPE_CHAR, 61},
{LLAMA_GRETYPE_RULE_REF, 7},
{LLAMA_GRETYPE_CHAR, 97},
},
{
{LLAMA_GRETYPE_RULE_REF, 5},
{LLAMA_GRETYPE_CHAR, 61},
{LLAMA_GRETYPE_RULE_REF, 7},
{LLAMA_GRETYPE_CHAR, 40},
},
{
{LLAMA_GRETYPE_RULE_REF, 5},
{LLAMA_GRETYPE_CHAR, 61},
{LLAMA_GRETYPE_RULE_REF, 7},
{LLAMA_GRETYPE_RULE_REF, 3},
{LLAMA_GRETYPE_CHAR, 48},
},
{
{LLAMA_GRETYPE_RULE_REF, 5},
{LLAMA_GRETYPE_CHAR, 61},
{LLAMA_GRETYPE_RULE_REF, 7},
{LLAMA_GRETYPE_RULE_REF, 3},
{LLAMA_GRETYPE_CHAR, 48},
},
{
{LLAMA_GRETYPE_RULE_REF, 5},
{LLAMA_GRETYPE_CHAR, 61},
{LLAMA_GRETYPE_RULE_REF, 7},
{LLAMA_GRETYPE_CHAR, 97},
}};
auto index = 0;
for (const llama_grammar_stack & stack : llama_grammar_get_stacks(grammar))
{
// compare stack to expected_stack
for (uint32_t i = 0; i < stack.size(); i++)
{
const llama_grammar_element * element = stack[i];
const llama_grammar_element & expected_element = expected_stacks[index][i];
// pretty print error message before asserting
if (expected_element.type != element->type || expected_element.value != element->value)
{
fprintf(stderr, "index: %d\n", index);
fprintf(stderr, "expected_element: %d, %u\n", expected_element.type, expected_element.value);
fprintf(stderr, "actual_element: %d, %u\n", element->type, element->value);
fprintf(stderr, "expected_element != actual_element\n");
}
assert(expected_element.type == element->type && expected_element.value == element->value);
}
index++;
}
std::vector<llama_grammar_candidate> next_candidates;
next_candidates.resize(23);
for (size_t i = 0; i < 23; ++i)
{
uint32_t *cp = new uint32_t[2]; // dynamically allocate memory for code_point
cp[0] = 37 + i;
cp[1] = 0;
next_candidates[i] = {i, cp, {}, 0};
}
std::vector<std::vector<std::pair<uint32_t, uint16_t>>> expected_reject = {
{
{0, 37},
{1, 38},
{2, 39},
{4, 41},
{5, 42},
{6, 43},
{7, 44},
{8, 45},
{9, 46},
{10, 47},
{11, 48},
{12, 49},
{13, 50},
{14, 51},
{15, 52},
{16, 53},
{17, 54},
{18, 55},
{19, 56},
{20, 57},
{21, 58},
{22, 59},
{23, 60},
},
{
{0, 37},
{1, 38},
{2, 39},
{3, 40},
{4, 41},
{5, 42},
{6, 43},
{7, 44},
{8, 45},
{9, 46},
{10, 47},
{21, 58},
{22, 59},
{23, 60},
},
{
{0, 37},
{1, 38},
{2, 39},
{3, 40},
{4, 41},
{5, 42},
{6, 43},
{7, 44},
{8, 45},
{9, 46},
{10, 47},
{21, 58},
{22, 59},
{23, 60},
},
{
{0, 37},
{1, 38},
{2, 39},
{3, 40},
{4, 41},
{5, 42},
{6, 43},
{7, 44},
{8, 45},
{9, 46},
{10, 47},
{11, 48},
{12, 49},
{13, 50},
{14, 51},
{15, 52},
{16, 53},
{17, 54},
{18, 55},
{19, 56},
{20, 57},
{21, 58},
{22, 59},
},
{
{0, 37},
{1, 38},
{2, 39},
{4, 41},
{5, 42},
{6, 43},
{7, 44},
{8, 45},
{9, 46},
{10, 47},
{11, 48},
{12, 49},
{13, 50},
{14, 51},
{15, 52},
{16, 53},
{17, 54},
{18, 55},
{19, 56},
{20, 57},
{21, 58},
{22, 59},
{23, 60},
},
{
{0, 37},
{1, 38},
{2, 39},
{3, 40},
{4, 41},
{5, 42},
{6, 43},
{7, 44},
{8, 45},
{9, 46},
{10, 47},
{21, 58},
{22, 59},
{23, 60},
},
{
{0, 37},
{1, 38},
{2, 39},
{3, 40},
{4, 41},
{5, 42},
{6, 43},
{7, 44},
{8, 45},
{9, 46},
{10, 47},
{21, 58},
{22, 59},
{23, 60},
},
{
{0, 37},
{1, 38},
{2, 39},
{3, 40},
{4, 41},
{5, 42},
{6, 43},
{7, 44},
{8, 45},
{9, 46},
{10, 47},
{11, 48},
{12, 49},
{13, 50},
{14, 51},
{15, 52},
{16, 53},
{17, 54},
{18, 55},
{19, 56},
{20, 57},
{21, 58},
{22, 59},
},
};
std::vector<llama_grammar_candidate> rejects = llama_grammar_reject_candidates_for_stack(llama_grammar_get_rules(grammar), llama_grammar_get_stacks(grammar)[0], next_candidates);
std::vector<std::vector<llama_grammar_candidate>> all_rejects;
for (std::size_t count = 0; count < llama_grammar_get_stacks(grammar).size(); ++count)
{
rejects = llama_grammar_reject_candidates_for_stack(llama_grammar_get_rules(grammar), llama_grammar_get_stacks(grammar)[count], next_candidates);
all_rejects.push_back(rejects);
}
index = 0;
for (auto rej : all_rejects)
{
for (uint32_t i = 0; i < rej.size(); i++)
{
auto element = rej[i];
auto expected_element = expected_reject[index][i];
assert(element.index == expected_element.first && *element.code_points == expected_element.second);
}
index++;
}
for (auto &candidate : next_candidates)
{
delete[] candidate.code_points;
candidate.code_points = nullptr;
}
llama_grammar_free_impl(grammar);
return 0;
}
Reference in a new issue View git blame Copy permalink