docs: byte-parity is precision-scoped - #208
Merged
Merged
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The finding
The golden matrix measured the byte-parity contract's real boundary:
fp32 artifacts reproduce byte-identically across hardware and
runtimes; quantized artifacts (int8/int4) are prefix-consistent but
stop-point-unstable — divergent outputs are exact prefixes of the
golden, differing only in where EOS wins a near-tie under
arch-level float differences.
The scoped contract
prefix-consistency; golden rows are reference outputs with
documented provenance
Same mechanism family as the beam-decode pathology: likelihood
ranking on near-uniform distributions.