feat(e6): Tashkeela-mix rung — registered + spec + data - #122
Merged
Conversation
Constant-budget register diversification: 8,000 classical Tashkeela units replace 8,000 news units (30,000 total; steps, schedule, optimizer, teacher identical to run-006/E5's control). Data: 120,000 decontaminated units at r5-units/tashkeela-scale.txt — stripped- normalized exact + 80-char-prefix decontamination against SadeedDiac-25 (48 dropped), deduped internally (17,036) and against the existing mix, seeded shuffle. Gate: adopt at <=4.5218; prediction 4.45-4.75. Launch gated on a free A100 slot.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
TODO.training-work/04 implemented: EXPERIMENTS.md E6 registered (gate before launch — adopt at <=4.5218, prediction 4.45-4.75), spec ara-diac-small-2-scale -> run-008-tashkeela-mix, and 120,000 decontaminated Tashkeela units on rababa-datasets (exact + 80-char-prefix decontamination against SadeedDiac-25, internal and cross-mix dedupe, seeded shuffle). Constant-budget design: 8k classical units replace 8k news units at 30k total — clean attribution of the register-diversification effect. Launch waits for a free A100 slot.