Refactor Where operator to avoid generation of NaN tensors #15
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
In pybuda where op is decomposed in a below way
In some cases, mask tensor(inputs[0]) may contain values (example : 3.4028234663852886e+38 -> torch.finfo(torch.float32).min -> highest representable value for a float32 ) other than 0s and 1s. subtraction of even very small values to this type of mask values should generate -infinity values which will lead to NaN tensor. To avoid that,
Fill values are getting replaced with ±1e4 without checking whether it is inf/-inf or not. To avoid that,