diff options
| author | Onur Gündüz <ogunduz@gmail.com> | 2026-06-23 01:07:04 +0200 |
|---|---|---|
| committer | Onur Gündüz <ogunduz@gmail.com> | 2026-06-23 01:07:04 +0200 |
| commit | 338ee4bf525175044df39add6eb559e5cc76edf6 (patch) | |
| tree | c822a844a03333827a47436374284213768ab8ba /tests/check_batch_parity.py | |
| parent | 99c090b8538b8b93f98e560b752c8c78bdfeb507 (diff) | |
| download | ffwd-338ee4bf525175044df39add6eb559e5cc76edf6.tar.gz ffwd-338ee4bf525175044df39add6eb559e5cc76edf6.zip | |
cuda: add bert ffn16 intermediate path
Diffstat (limited to 'tests/check_batch_parity.py')
| -rwxr-xr-x | tests/check_batch_parity.py | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/tests/check_batch_parity.py b/tests/check_batch_parity.py index 3b02a34..fa94540 100755 --- a/tests/check_batch_parity.py +++ b/tests/check_batch_parity.py @@ -140,7 +140,7 @@ def main(): choices=(0, 8)) ap.add_argument("--mlx-quant-group-size", "--mlx-quantize-group-size", dest="mlx_quantize_group_size", type=int, default=64) - ap.add_argument("--cuda-weight-dtype", choices=("f32", "bf16"), default=None, + ap.add_argument("--cuda-weight-dtype", choices=("f32", "bf16", "f16"), default=None, help="force CUDA weight storage dtype (default: snapshot dtype)") ap.add_argument("texts", nargs="*") args = ap.parse_args() |
