zhangyiss/mlx
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx.git synced 2025-12-16 01:49:05 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

add fp8 e4m3 converters

Browse Source
This commit is contained in:
Awni Hannun
2025-10-20 08:49:35 -07:00
parent 9f0ba3ddf1
commit ec4c85d786
16 changed files with 312 additions and 111 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
tests/ops_tests.cpp
View File

@@ -4032,3 +4032,25 @@ TEST_CASE("test conv_transpose3d with output_padding") {
{1, 2, 4, 4, 1});
CHECK(array_equal(out, expected).item<bool>());
}
TEST_CASE("test fp8 conversion") {
for (auto t : {float32, float16, bfloat16}) {
array in({-1.125, -1.0, 0.0, 1.0, 1.125, 4.5, 448.0}, t);
auto in_fp8 = to_fp8(in);
auto out = from_fp8(in_fp8, t);
}
array in({-1.125, -1.0, 0.0, 1.0, 1.125, 4.5, 448.0});
array noisy_in({-1.135, -1.01, 0.0001, 1.01, 1.135, 4.6, 447.0});
auto in_fp8 = to_fp8(noisy_in);
auto out = from_fp8(in_fp8, float32);
CHECK(array_equal(out, in).item<bool>());
// Overflow
in = array({-600.0, 600.0});
in_fp8 = to_fp8(in);
out = from_fp8(in_fp8, float32);
auto expected = array({NAN, NAN});
CHECK(array_equal(out, expected, true).item<bool>());
}