zooL4nD3r demo

Translate any passage into the discourse style of one of 961 learned communities, while keeping the meaning intact.

Powered by RiverRider/zooL4nD3r-v0.1: a 14.6 M-parameter side-channel adapter for Qwen/Qwen2.5-7B, trained to discriminate between fine-grained discourse-mode clusters extracted from a curated multi-domain corpus (dialog, QA, philosophy, semiotics).

How it works

1. Pick a target community from the dropdown (961 options, search-friendly).
2. Paste a passage.
3. Submit. The instruct backbone rewrites the passage in the target community's style; the adapter scores how close the rewrite landed to the target centroid, and how much meaning was preserved vs the source.

Limitations

• This is a single-shot rewrite, not the streaming token-tape steering UI from the full zooL4nD3r runtime.
• Cold-start can take 3-4 min while community centroids are computed once.
• Specialized for long-form discourse; short or single-sentence inputs will rewrite less reliably.
• English-only training corpus; cross-lingual transfer is partial.

Source code

The thin client app is open source at github.com/space-bacon/zooL4nD3r-public under Apache-2.0.

The adapter weights are open at RiverRider/zooL4nD3r-v0.1 under Apache-2.0.

The full streaming runtime, multi-adapter orchestration, and translation engine are proprietary.