Skip to content

Support Half/BFloat16 in log_softmax #7826

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
swolchok merged 1 commit into from
Jan 22, 2025
Merged

Support Half/BFloat16 in log_softmax #7826

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion kernels/portable/cpu/op_log_softmax.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,7 +42,7 @@ Tensor& log_softmax_out(
// Adjust for negative dim
dim = dim < 0 ? dim + nonzero_dim(in) : dim;

ET_SWITCH_FLOAT_TYPES(
ET_SWITCH_FLOATHBF16_TYPES(
in.scalar_type(), ctx, "_log_softmax.out", CTYPE, [&]() {
const CTYPE* const in_data = in.const_data_ptr<CTYPE>();
CTYPE* const out_data = out.mutable_data_ptr<CTYPE>();
Expand Down
18 changes: 12 additions & 6 deletions kernels/test/op_log_softmax_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,7 +62,15 @@ class OpLogSoftmaxOutTest : public OperatorTest {
});
// clang-format on

EXPECT_TENSOR_CLOSE(out, expected);
if constexpr (DTYPE == ScalarType::BFloat16) {
EXPECT_TENSOR_CLOSE_WITH_TOL(
out,
expected,
1e-2,
executorch::runtime::testing::internal::kDefaultAtol);
} else {
EXPECT_TENSOR_CLOSE(out, expected);
}
}
};

Expand All @@ -88,11 +96,9 @@ TEST_F(OpLogSoftmaxOutTest, AllDtypesSupported) {
GTEST_SKIP() << "This kernel does not support dtype double";
}

test_dtype<float, ScalarType::Float>();
test_dtype<double, ScalarType::Double>();
// TODO: Also add tests for half, complex, quantized, and other types. Easiest
// way to do that would be to make TensorFactory support zeros() and ones()
// for those types.
#define TEST_ENTRY(ctype, dtype) test_dtype<ctype, ScalarType::dtype>();
ET_FORALL_FLOATHBF16_TYPES(TEST_ENTRY)
#undef TEST_ENTRY
}

TEST_F(OpLogSoftmaxOutTest, MismatchedDimensionsDies) {
Expand Down
Loading