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

test-backend-ops : test scale parameter of ggml_soft_max_ext

Browse source
This commit is contained in:
Jared Van Bortel 2024-01-24 16:12:42 -05:00
parent 2852902eda
commit 1450966071
1 changed files with 8 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

12
tests/test-backend-ops.cpp
View file

@ -1056,18 +1056,20 @@ struct test_diag_mask_inf : public test_case {
struct test_soft_max : public test_case {
const ggml_type type;
const std::array<int64_t, 4> ne;
const float scale;
std::string vars() override {
return VARS_TO_STR2(type, ne);
return VARS_TO_STR3(type, ne, scale);
}
test_soft_max(ggml_type type = GGML_TYPE_F32,
std::array<int64_t, 4> ne = {10, 10, 10, 10})
: type(type), ne(ne) {}
std::array<int64_t, 4> ne = {10, 10, 10, 10},
float scale = 1.0f)
: type(type), ne(ne), scale(scale) {}
ggml_tensor * build_graph(ggml_context * ctx) override {
ggml_tensor * a = ggml_new_tensor(ctx, type, 4, ne.data());
ggml_tensor * out = ggml_soft_max(ctx, a);
ggml_tensor * out = ggml_soft_max_ext(ctx, a, nullptr, scale);
return out;
}
};
@ -1825,6 +1827,8 @@ static bool test_backend(ggml_backend_t backend, test_mode mode, const char * op
exponent <<= 1;
}
test_cases.emplace_back(new test_soft_max(GGML_TYPE_F32, {16, 16, 1, 1}, 0.1f));
for (ggml_type type : {GGML_TYPE_F32, GGML_TYPE_F16}) {
test_cases.emplace_back(new test_rope(type, {128, 32, 10, 1}, 128, 0, 512)); // llama 7B
test_cases.emplace_back(new test_rope(type, {128, 40, 10, 1}, 128, 0, 512)); // llama 13B