Preserve FP32 coordinates under BF16 in tomography and for INRs - #280
Open
ccasert wants to merge 1 commit into
Open
Preserve FP32 coordinates under BF16 in tomography and for INRs#280ccasert wants to merge 1 commit into
ccasert wants to merge 1 commit into
Conversation
cedriclim1
self-requested a review
August 20, 2026 21:02
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What problem this PR addreseses
This PR fixes BF16 instability in INRs by keeping rays and sampler coordinates (for kplanes) in FP32.
Coordinates are normalized to
[-1, 1]. At high enough resolution, training INRs with BF16 can cause instability because the spacing between points can become less than BF16 can represent, so adjacent samples can collapse to the same value.It also applies BF16 consistently during training, validation, and pretraining for tomography; previously reconstruction used BF16 only during validation.
What should the reviewer(s) do
Review the FP32 and BF16 coordinate behavior.
use_bfloat16).