feat(distill): ara-diac-tiny spec — the browser/Worker tier probe - #63
Merged
Conversation
~25-45M-param student (d384, enc6/dec2, byte vocab), single-variable vs run-002: same r6 teacher, corpus, frozen labels. Adds config-only student init for widths with no pretrained backbone. Gate: no collapse; DER <= 5.0 windowed would ship as the tiny tier.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 browser/Worker tier needs ~30-80MB artifacts (current smallest
int8 = 202MB; a Cloudflare Worker isolate caps at 128MB). This adds
the ara-diac-tiny spec: a d384/enc6/dec2 byte student (~25-45M params
-> ~30-45MB int8), trained on the identical r6 teacher labels and
corpus as ara-diac-small run-002, so the frontier table gains a
single-variable capacity rung. Also adds student_config support
(random init at custom width - no pretrained backbone exists at this
size; dense CE supervision is the mitigation, noted on the spec).
Launch is a detached Modal run; results land on
rababa-checkpoints:/rababa_arabic_distill_tiny/run-001.