Is float16 enabled?

https://github.com/bytedance/Protenix/blob/128b33bede180df9737f8be245f7ba5e08c03db1/protenix/openfold_local/model/triangular_multiplicative_update.py#L597

I could be missing something but don't you need to un-scale the output of the matmul here? Otherwise this will give different results.

Start at protenix/openfold_local/model/triangular_multiplicative_update.py around line 597 and trace the float16 matmul path. Check the surrounding scaling logic and any existing tests or callers for this operation. Done means confirming whether the output requires un-scaling and documenting or correcting the behavior with appropriate test coverage.

Written by the indexing model from the issue text.