tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-26 19:55:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat: add GGML_UNARY_OP_ARGMAX Metal kernel (ggml/1019)

Browse Source 
* implemented argmax kernel

* tpig -> tgpig

* change to strides

* contiguous assertions

* kernel working and tested

* argmax simd parallel implementation

* added 2 new tests for argmax in test-backend-ops

* cosmit

* added 3 tests cases for perf eval

* add test_argmax in make_test_cases_perf

* Update test-backend-ops.cpp

Co-authored-by: Diego Devesa <slarengh@gmail.com>

---------

Co-authored-by: Diego Devesa <slarengh@gmail.com>
This commit is contained in:
PAB
2024-12-02 19:27:24 +01:00
committed by Georgi Gerganov
parent 667d70d170
commit efb6ae9630
3 changed files with 92 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@ -3460,14 +3460,15 @@ 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_argmax());
test_cases.emplace_back(new test_argmax(GGML_TYPE_F32, {32, 1, 1, 1}));
test_cases.emplace_back(new test_argmax(GGML_TYPE_F32, {100, 10, 1, 1}));
test_cases.emplace_back(new test_argmax(GGML_TYPE_F32, {1024, 10, 1, 1}));
test_cases.emplace_back(new test_argmax(GGML_TYPE_F32, {2000, 10, 1, 1}));
test_cases.emplace_back(new test_count_equal());
test_cases.emplace_back(new test_argmax(GGML_TYPE_F32, {32, 1, 1, 1}));
test_cases.emplace_back(new test_argmax(GGML_TYPE_F32, {100, 10, 1, 1}));
test_cases.emplace_back(new test_argmax(GGML_TYPE_F32, {1024, 10, 1, 1}));
test_cases.emplace_back(new test_argmax(GGML_TYPE_F32, {1024, 12, 1, 1}));
test_cases.emplace_back(new test_argmax(GGML_TYPE_F32, {2000, 10, 1, 1}));
test_cases.emplace_back(new test_argmax(GGML_TYPE_F32, {5438, 3, 1, 1}));
for (int ne3 : {1, 3}) { // CUDA backward pass only supports ne3 == 1
test_cases.emplace_back(new test_repeat(GGML_TYPE_F32, {10, 5, 4, ne3}, {1, 1, 1, 1}));
test_cases.emplace_back(new test_repeat(GGML_TYPE_F32, {10, 5, 4, ne3}, {2, 1, 1, 1}));