feat(playground): speculative vocalization tier (2.1 quality, int4 drafts) - #184
Merged
Merged
Conversation
…afts) Third vocalization option: ara-diac-layerdrop-1.0-int4 drafts, ara-diac-small-2.1-int8 verifies in one pass per block - measured acceptance 0.9886 at 8.85 tokens per verifier pass (RESULTS.md 2026-09-11), output-preserving by construction. Status line reports the acceptance after each run. Requires interscript ^5.4.0 (SpeculativeModel). Live e2e (SPEC_E2E=1): downloads both models in a fresh profile, vocalizes, and shows the readout (1.6m). Also: stale CLI-name test expectations (interscript-ts -> interscript, pre-existing red on main), dead frontmatter var + unused SITE (lint), playground.astro added to .prettierignore (hand-wrapped prose).
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Adds the speculative vocalization tier to the playground - the runtime lever measured in the DeepSeek-V4.1-Flash learnings pass (rababa TODO.impl/01).
What users get. A third vocalization option: "Arabic 2.1 + speculative" - ara-diac-layerdrop-1.0-int4 (95 MB) drafts, ara-diac-small-2.1-int8 (264 MB) verifies every block in one pass. Same 2.1 outputs (greedy verification is output-preserving), at ~9x fewer verifier decode invocations: measured acceptance 0.9886, 8.85 tokens per verifier pass (interscript-ml RESULTS.md, 2026-09-11). After each run the status line reports the acceptance readout.
Dependency. interscript ^5.4.0 (published; exports imf.SpeculativeModel).
Verification.
Incidentals (pre-existing red on main, fixed here): stale CLI-name test expectations (interscript-ts vs the shipped interscript CLI), unused frontmatter var, unused SITE constant, playground.astro added to .prettierignore per the hand-wrapped-prose convention.