zhangyiss/mlx
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx.git synced 2025-10-18 07:18:29 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

enable admm low-precision cpu (#2661)

Browse Source
This commit is contained in:
Awni Hannun
2025-10-10 09:50:54 -07:00
committed by GitHub
parent f37389d100
commit 380aeb58ae
2 changed files with 9 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
mlx/backend/cpu/matmul.cpp
View File

@@ -131,10 +131,6 @@ void Matmul::eval_cpu(const std::vector<array>& inputs, array& out) {
}
void AddMM::eval_cpu(const std::vector<array>& inputs, array& out) {
if (out.dtype() != float32) {
throw std::runtime_error(
"[AddMM::eval_cpu] Currently only supports float32.");
}
if (out.size() == 0) {
out.set_data(allocator::malloc(out.nbytes()));
return;

9
python/tests/test_blas.py
View File

@@ -712,6 +712,15 @@ class TestBlas(mlx_tests.MLXTestCase):
expected = beta * c + alpha * (a @ b)
self.assertTrue(mx.allclose(expected, out))
# Test half precision
for t, tol in [(mx.float16, 1e-3), (mx.bfloat16, 1e-2)]:
c = mx.ones((32, 32)).astype(t)
a = mx.random.uniform(shape=(32, 32)).astype(t)
b = mx.random.uniform(shape=(32, 32)).astype(t)
out = mx.addmm(c, a, b)
expected = a @ b + c
self.assertTrue(mx.allclose(out, expected, rtol=tol, atol=tol))
def test_addmm_grad(self):
def make_ref_addmm(alpha, beta):
return lambda c, a, b: alpha * (a @ b) + beta * c