tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-26 11:45:21 +00:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: batched+noncont MMQ, refactor bs>1 MoE code (#13199)

Browse Source
This commit is contained in:
Johannes Gäßler
2025-04-30 23:12:59 +02:00
committed by GitHub
parent 6f67cf1f48
commit e1e8e0991f
9 changed files with 869 additions and 440 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
tests/test-backend-ops.cpp
View File

@ -4184,6 +4184,11 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {2, 3}, {1, 1}, {0, 2, 1, 3}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {2, 3}, {1, 1}, {0, 1, 3, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {2, 3}, {1, 1}, {0, 3, 2, 1}));
// test cases with large ne00/ne10 to cover stream-k fixup
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 1024, {3, 2}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 8, 1024, {3, 2}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 1024, {3, 2}, {1, 1}));
}
}
for (ggml_type type_a : other_types) {