Licensing and attribution
Qev materials
Copyright 2026 Qiqian Fu and Qev contributors.
Qev's code, documentation, original SVG illustrations, example requests, and generated prediction results are provided under the Apache License 2.0. The Qev adaptation weights published at AustinFu/Qev-9B, including its LoRA adapter, decision head, and interaction gate, use the same license. Contributions are accepted under these terms unless explicitly stated otherwise.
Third-party materials retain the terms listed below. The Qev license does not relicense training datasets, external services, or third-party recordings.
Included and adapted materials
| Material | Source and copyright | License and scope |
|---|---|---|
| Kev conventions and adapted implementation | Jared Palmer's Kev, Copyright 2026 Jared Palmer | Apache-2.0, original notice retained. Qev adapts delimiter and rendering conventions, LoRA targets, and cache forking. Modified source files identify the Qev adaptation. |
| Qwen models and tokenizers | Qwen3.5-9B-Base and Qwen3.5-2B-Base, Copyright 2026 Alibaba Cloud | Apache-2.0. Tokenizers accompany the corresponding Qev checkpoints; base weights are downloaded separately. |
| JevBench public evaluation tasks | JevBench, Copyright 2026 Florian Standhartinger and contributors | MIT, with upstream third-party notices. Tasks are downloaded by the optional preparation script. The original license, third-party notice, and per-question provenance accompany the prepared dataset. |
| Crafter visuals in the gameplay recording | Crafter, Copyright 2021 Danijar Hafner | MIT. The gameplay recording contains the environment's visual assets; the recording and Qev decision overlay were produced for this project. |
The reviewed Kev version is 557598f; the evaluation data use JevBench v1.4.2. The 231 downloaded public tasks carry MIT in their individual provenance records; upstream notices describe the scope of other JevBench materials and evaluated services.
The README layout takes inspiration from JevAny. Qev's SVG illustrations and chart code are original. The GIF and MP4 demos are this project's recordings of 9B research models. The Crafter research integration used JevAny's environment adapters; this source package does not vendor those adapters or JevAny's bundled recordings.
Dependencies installed separately
Dependencies are installed from their own distributions and retain their licenses:
- PyTorch: BSD-3-Clause, plus its bundled third-party notices.
- Transformers, PEFT, Accelerate, Safetensors, Hugging Face Hub, and optional Apache Arrow: Apache-2.0.
- Development tools pytest and Matplotlib retain their MIT and Matplotlib/PSF-based license terms, respectively.
Their source code and dependency distributions are not vendored in Qev's Python package. This list identifies direct dependencies; their own packages include notices for transitive components.
Training data
The complete research training mixture is not distributed here. Public weight availability does not make the training data available under Apache-2.0. Any future dataset release needs its own source-specific permissions and attribution. The Wikipedia-derived training subset, for example, carries CC BY-SA attribution where applicable.
The small JSONL requests in examples/ were written for Qev and are covered by its Apache-2.0 license. The benchmark result tables and prediction files contain Qev's recorded outputs, not the original benchmark question text.