Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 2 additions & 4 deletions CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -3,14 +3,12 @@
Versions follow [semantic versioning](https://semver.org/). Each release pins the model revisions it loads
(`jev_style/models.py`), so upgrading the package is what changes the weights you get.

## Unreleased

Documentation only. The README is restructured the way the larger decision-model repositories present themselves: quick start and the reasons to use it first, one model table with a recommendation, a comparison with hosted Jev, Kev and Laya, then the agent tooling, and new sections on how the model works, its limitations, where it is used, a roadmap and a citation. The identity, training-source and service-status statements move to an "About the project" section at the end, unchanged in substance. It carries the 0.6.4 README content: the Decision Index 0.3 board entry (2B v4 Public 44.37 / Full 34.53 / Breadth 41.75, Vision 57.55 / 51.94; 0.8B v3 17.0 / 12.5), the hosted API's model, context, input limit and 429 / `Retry-After` behaviour, install commands with the `[torch]` / `[mlx]` extras and the `mlx-lm` pin, the key rule, both error-body shapes, the local server's Host / Origin check, the download and `--verify` integrity checks, `noul` / `choice` examples in place of the urgency `score` ones, and GitHub Discussions and `SECURITY.md` for questions and security reports. The outdated 2B v4 banner is no longer shown. Added `README.zh-CN.md`, `CITATION.cff`, a logo and a social-preview image under `docs/assets/`. Runtime behavior, dependencies, model revisions and weights are unchanged.

## 0.6.4 (2026-11-05)

Fixes from the 2026-11-05 review of the package. Model revisions and weights are unchanged from 0.6.0.

- **README.** The README is restructured the way the larger decision-model repositories present themselves: quick start and the reasons to use it first, one model table with a recommendation, a comparison with hosted Jev, Kev and Laya, then the agent tooling, and new sections on how the model works, its limitations, where it is used, a roadmap and a citation. The identity, training-source and service-status statements move to an "About the project" section at the end, unchanged in substance. It carries the 0.6.4 README content: the Decision Index 0.3 board entry (2B v4 Public 44.37 / Full 34.53 / Breadth 41.75, Vision 57.55 / 51.94; 0.8B v3 17.0 / 12.5), the hosted API's model, context, input limit and 429 / `Retry-After` behaviour, install commands with the `[torch]` / `[mlx]` extras and the `mlx-lm` pin, the key rule, both error-body shapes, the local server's Host / Origin check, the download and `--verify` integrity checks, `noul` / `choice` examples in place of the urgency `score` ones, and GitHub Discussions and `SECURITY.md` for questions and security reports. The outdated 2B v4 banner is no longer shown. Added `README.zh-CN.md`, `CITATION.cff`, a logo and a social-preview image under `docs/assets/`. Runtime behavior, dependencies, model revisions and weights are unchanged.

- **Skills and listings.** The `jev-style-serve` skill no longer suggests piping an install script into `sh`; it
points to `brew install uv` / `pipx install uv` (skill scanners graded the old line as a supply-chain risk). The
guard skill's description says "recursive deletes, force-pushes or key exfiltration" instead of literal commands.
Expand Down
Loading