Fix mantissa underflow in block floating point quantization#21
Open
Shayanthn wants to merge 1 commit into
Open
Fix mantissa underflow in block floating point quantization#21Shayanthn wants to merge 1 commit into
Shayanthn wants to merge 1 commit into
Conversation
Fixes ma3mool#15 After dividing mantissa by power_exp_diff during shared exponent alignment, values can become less than 1.0 which leads to incorrect exponent reconstruction. This fix clamps underflowed mantissa values to zero to maintain valid block floating point representation. Applied fix to both quant_bfloat_py and quant_bfloat_meta_py functions.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Fixes #15
Problem
In block floating point quantization, after dividing mantissa by
power_exp_diffduring shared exponent alignment, values can become less than 1.0. This causes incorrect exponent reconstruction when the tensor is later processed.Solution
Added mantissa underflow handling by clamping values below 1.0 to zero, maintaining valid block floating point representation.
Changes
quant_bfloat_py: Added underflow check after mantissa shiftingquant_bfloat_meta_py: Applied same fix for consistency