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

metal : small-batch mat-mul kernels (#10581)

Browse source 
* metal : small-batch mat-mul kernels

ggml-ci

* metal : add rest of types

ggml-ci

* metal : final adjustments

ggml-ci

* metal : add comments

ggml-ci
This commit is contained in:
Georgi Gerganov 2024-12-03 11:52:33 +02:00 • committed by GitHub
parent 515d4e5372
commit 0115df2f65
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 709 additions and 29 deletions
Download patch file Download diff file Expand all files Collapse all files

13
tests/test-backend-ops.cpp
View file

@ -3572,6 +3572,19 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_rwkv_wkv6(GGML_TYPE_F32, 32, 64, 32, 4));
test_cases.emplace_back(new test_rwkv_wkv6(GGML_TYPE_F32, 32, 64, 128, 4));
for (int i = 1; i < 9; ++i) {
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_F16, GGML_TYPE_F32, 16, i, 256, { 1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q4_0, GGML_TYPE_F32, 16, i, 256, { 1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q4_1, GGML_TYPE_F32, 16, i, 256, { 1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q5_0, GGML_TYPE_F32, 16, i, 256, { 1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q5_1, GGML_TYPE_F32, 16, i, 256, { 1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q8_0, GGML_TYPE_F32, 16, i, 256, { 1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q4_K, GGML_TYPE_F32, 16, i, 256, { 1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q5_K, GGML_TYPE_F32, 16, i, 256, { 1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_Q6_K, GGML_TYPE_F32, 16, i, 256, { 1, 1}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(GGML_TYPE_IQ4_NL, GGML_TYPE_F32, 16, i, 256, { 1, 1}, {1, 1}));
}
#if 1
for (ggml_type type_a : base_types) {
for (ggml_type type_b : {GGML_TYPE_F32, GGML_TYPE_F16}) {