vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge branch 'master' into compilade/cuda-tq2_0

Browse source
This commit is contained in:
Francis Couture-Harpin 2025-01-09 13:02:09 -05:00
commit 983aa09b5c
217 changed files with 19585 additions and 15473 deletions
Download patch file Download diff file Expand all files Collapse all files

14
tests/test-backend-ops.cpp
View file

@ -3940,7 +3940,7 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_perf() {
test_cases.emplace_back(new test_argmax(GGML_TYPE_F32, {1024, 10, 1, 1}));
test_cases.emplace_back(new test_argmax(GGML_TYPE_F32, {32000, 512, 1, 1}));
for (int bs : {1, 512}) {
for (int bs : {1, 2, 3, 4, 5, 8, 512}) {
for (ggml_type type_a : all_types) {
for (ggml_type type_b : {GGML_TYPE_F32}) {
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 4096, bs, 14336, {1, 1}, {1, 1}));
@ -3948,6 +3948,18 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_perf() {
}
}
for (int K : {3, 5}) {
for (int IC : {256, 2560}) {
for (int IW_IH : {32, 64, 256}) {
if (IC == 2560 && IW_IH == 256) {
// too big
continue;
}
test_cases.emplace_back(new test_im2col(GGML_TYPE_F32, GGML_TYPE_F16, GGML_TYPE_F32, {IW_IH, IW_IH, IC, 1}, {K, K, IC, 1}, 1, 1, 1, 1, 1, 1, true));
}
}
}
return test_cases;
}