Repository navigation
Add HISTORY, CONTRIBUTING and CODE_OF_CONDUCT; stop tracking agent files - #11
Merged
Merged
Conversation
…nd AGENTS.md; remove local paths from public docs
…; mark Session 5 in progress
Adds the scripts that drove Parentheses 1.0-240M Sessions 3 to 7 and the Session 7.5 calibrated SFT rebuild: - Session builders and launchers: S4 STEM, S5 textbooks, S6 anneal, S7 SFT, S7.5 calibrated SFT (11-language replay, per-bucket token-share gate, explicit lang map, long-think rescue), plus supervised runners. - Eval: 60-probe adversarial suite and scorer, 11-language BPB and BPB gate, blind rubric, MQAR probe. - Ops: checkpoint manager with SHA256 archive, telemetry watchdog, sentinel. - Model: opt-in gated delta-rule state update and beta_mode for slab_v4 (defaults unchanged), small A/B preset, delta parity test. - Loaders: dual_loader secondary_per_16 replay ratio; packer reads the `language` field.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
docs/HISTORY.md: the story of parentheses from the 2024 transformer (0.1) to 1.0-240M, with a parameter-growth chart. Only 0.1, 0.9 and 1.0 were names used at the time; the 2026 stages are labelled 0.2 to 0.8 after the fact, and each one is listed with the name it actually ran under. It says plainly that the 0.9 weights are non-commercial because some of their instruction data was (Camel-AI, SciQ, No Robots, and Karya with no licence stated), and how that led to the 2026-11-02 licence rule. ROADMAP and README link to it.CONTRIBUTING.md: open problems (licensed Dravidian text, native-speaker checks, setup reports, bugs), the data licence rule, code basics, and the contact address mail@uncoalesced.com.CODE_OF_CONDUCT.md: why the model is uncensored (no refusal training, no guard models, no content filter, still honest about what it doesn't know) and how contributors treat each other, with reports to mail@uncoalesced.com.CLAUDE.mdandAGENTS.mdare no longer tracked. Both were already listed in.gitignoreand contained personal names and local paths; the files stay on the maintainer's machine. Public references toAGENTS.mdwere reworded.E:\/I:\paths andfile:///links removed fromMODEL_CARD.mdand three files indocs/, and the chart script no longer hardcodes a personal font path (it usesPARENTHESES_FONTor the Windows font folder).LICENSE_POLICY.md(and its copy inscripts/) updated to the 2026-11-02 rule that the docs already describe.Each new document was checked by a separate reviewer against the source files, and its fixes are applied.
Not in this PR
Many scripts and
DATA_LEDGER.mdstill contain hardcoded local paths,GATES.mdis still tracked, andMODEL_CARD.md(the 120M card) still states Apache 2.0 for weights.