zooL4nD3r-public

Public assets for the zooL4nD3r v0.1 release: model card, demo Space source, and architecture docs.

The full streaming runtime, multi-adapter zoo, translation engine, and chat UI are proprietary and live in a separate private repo.

Links

• Model: RiverRider/zooL4nD3r-v0.1 (Apache-2.0)
• Live demo: RiverRider/zooL4nD3r-demo (ZeroGPU)
• Eval report: docs/training.md

What's in this repo

space/             # Gradio ZeroGPU demo source (= the live Space's app.py)
docs/
  model_card.md    # = RiverRider/zooL4nD3r-v0.1 README
  training.md      # full training + eval report (vs srt-adapter-v22c_a050)
  architecture.md  # high-level system + IP boundary
  api.md           # streaming SSE schema (for v0.2 streaming runtime)
LICENSE            # Apache-2.0
README.md

What's NOT in this repo

The streaming chat runtime, multi-adapter zoo, translation engine, slider- based generation steering, FastAPI / Modal serving glue, and corpus build pipeline are proprietary and live in a separate private repo (space-bacon/zooL4nD3r). The live demo Space implements only single-shot cross-community translation, which is enough to showcase the v0.1 adapter's strength on its own.

Quickstart

Run the demo locally (needs a CUDA GPU and ~20 GB free for the Qwen/Qwen2.5-7B-Instruct backbone download):

cd space
pip install -r requirements.txt
python app.py

First launch downloads the backbone (~15 GB) and builds 961 community centroids (~3-4 min on an A10G). Subsequent launches use the cache.

Citation

@misc{zoolander-v0.1,
  title  = {zooL4nD3r v0.1: A community-discourse side-channel adapter for Qwen2.5-7B},
  author = {RiverRider},
  year   = {2026},
  url    = {https://huggingface.co/RiverRider/zooL4nD3r-v0.1}
}

License

Apache-2.0. See LICENSE.