CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-27 08:21:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' into compilade/test-model-random

Browse Source
This commit is contained in:
Francis Couture-Harpin
2025-06-26 17:23:16 -04:00
parent ccb2bb9988 8846aace49
commit 7c3f9c226f
142 changed files with 9419 additions and 6305 deletions
Download Patch File Download Diff File Expand all files Collapse all files

100
tests/test-backend-ops.cpp
View File

@@ -2725,6 +2725,35 @@ struct test_conv_transpose_1d : public test_case {
}
};
// GGML_OP_CONV_TRANSPOSE_2D
struct test_conv_transpose_2d : public test_case {
const std::array<int64_t, 4> ne_input;
const std::array<int64_t, 4> ne_kernel;
const int stride;
std::string vars() override {
return VARS_TO_STR3(ne_input, ne_kernel, stride);
}
test_conv_transpose_2d(std::array<int64_t, 4> ne_input = {10, 10, 3, 1}, // [input_width, input_height, input_channels, 1]
std::array<int64_t, 4> ne_kernel = {3, 3, 3, 1}, // [kernel_width, kernel_height, input_channels, 1]
int stride = 1)
: ne_input(ne_input), ne_kernel(ne_kernel), stride(stride){}
ggml_tensor * build_graph(ggml_context * ctx) override {
ggml_tensor * input = ggml_new_tensor(ctx, GGML_TYPE_F32, 4, ne_input.data());
ggml_set_name(input, "input");
ggml_tensor * kernel = ggml_new_tensor(ctx, GGML_TYPE_F16, 4, ne_kernel.data());
ggml_set_name(kernel, "kernel");
ggml_tensor * out = ggml_conv_transpose_2d_p0(ctx, kernel, input, stride);
ggml_set_name(out, "out");
return out;
}
};
// GGML_OP_IM2COL
struct test_im2col : public test_case {
const ggml_type type_input;
@@ -4050,6 +4079,9 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_conv_transpose_1d({3,2,1,1}, {3,1,2,1}, 1, 0, 1));
test_cases.emplace_back(new test_conv_transpose_1d({2,1,1,1}, {3,1,1,1}, 1, 0, 1));
test_cases.emplace_back(new test_conv_transpose_2d({3, 2, 3, 1}, {2, 2, 1, 3}, 1));
test_cases.emplace_back(new test_conv_transpose_2d({10, 10, 9, 1}, {3, 3, 1, 9}, 2));
test_cases.emplace_back(new test_count_equal(GGML_TYPE_F32, {4, 500, 1, 1}));
test_cases.emplace_back(new test_count_equal(GGML_TYPE_F32, {4, 5000, 1, 1}));
@@ -4220,39 +4252,45 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
#if 1
for (ggml_type type_a : base_types) {
for (ggml_type type_b : {GGML_TYPE_F32, GGML_TYPE_F16}) {
// test cases without permutation
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {1, 1}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {1, 1}, {1, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {3, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {3, 1}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {3, 2}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {3, 2}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {3, 2}, {1, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {3, 2}, {2, 2}));
std::vector<int> ks = { 256 };
if (ggml_blck_size(type_a) == 1) {
ks.push_back(4);
}
for (auto k : ks) {
// test cases without permutation
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {1, 1}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {1, 1}, {1, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {3, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {3, 1}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {3, 2}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {3, 2}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {3, 2}, {1, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {3, 2}, {2, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {1, 1}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {1, 1}, {1, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {3, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {3, 1}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {3, 2}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {3, 2}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {3, 2}, {1, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 256, {3, 2}, {2, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {1, 1}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {1, 1}, {1, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {3, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {3, 1}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {3, 2}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {3, 2}, {2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {3, 2}, {1, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {3, 2}, {2, 2}));
// test cases with permutation
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {2, 3}, {1, 1}, {0, 2, 1, 3}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {2, 3}, {1, 1}, {0, 1, 3, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 256, {2, 3}, {1, 1}, {0, 3, 2, 1}));
// test cases with permutation
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {2, 3}, {1, 1}, {0, 2, 1, 3}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {2, 3}, {1, 1}, {0, 1, 3, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, k, {2, 3}, {1, 1}, {0, 3, 2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 8, 256, {2, 3}, {1, 1}, {0, 2, 1, 3}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 8, 256, {2, 3}, {1, 1}, {0, 1, 3, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 8, 256, {2, 3}, {1, 1}, {0, 3, 2, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 8, k, {2, 3}, {1, 1}, {0, 2, 1, 3}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 8, k, {2, 3}, {1, 1}, {0, 1, 3, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 8, k, {2, 3}, {1, 1}, {0, 3, 2, 1}));
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.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {2, 3}, {1, 1}, {0, 2, 1, 3}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {2, 3}, {1, 1}, {0, 1, 3, 2}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, k, {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}));
@@ -4618,6 +4656,10 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_perf() {
test_cases.emplace_back(new test_conv_2d_dw({512, 512, 256, 1}, {3, 3, 1, 256}, 1, 1, 1, false));
test_cases.emplace_back(new test_conv_2d_dw({512, 512, 256, 1}, {3, 3, 1, 256}, 1, 1, 1, true));
test_cases.emplace_back(new test_conv_transpose_2d({256, 256, 256, 1}, {3, 3, 16, 256}, 1));
test_cases.emplace_back(new test_mean(GGML_TYPE_F32, {256, 256, 3, 1}));
return test_cases;
}

1
tests/test-model-random.cpp
View File

@@ -573,6 +573,7 @@ struct model_variant {
}
} break;
case LLM_ARCH_GEMMA3:
case LLM_ARCH_GEMMA3N:
case LLM_ARCH_STARCODER2:
break;
case LLM_ARCH_MAMBA: