Repository navigation
feat: advise source-proved unamortized activation packing (PS6141) - #1015
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Related to issue #835; broader codec/native coverage and evidence-based issue audit remain.
Summary
Evidence and limits
Authentic retained Q8_0 encoder and scalar-half source support is separated from synthetic/MIXED complete source fixtures. Actual retained float-activation/packed-weight row kernels are negative or opaque, not invented historical activation-positive evidence. No numerical equivalence, reciprocal/approximate rewrite, portable transform or speedup is claimed.
The documented Apple M2 Pro owner observations are attributed, not locally reproduced: established median 182032 ns/op versus packed median 208481 ns/op, about 0.87x and one extra allocation. Benchmark the complete conversion/consumer boundary at the relevant target, shape, precision and lifetime.
Validation
Further acceptance requires exact-head CI and root review; this PR does not establish full native/model or general codec coverage.