Benchmarks & research

Compass by TDCC

TDCC's Compass is an experimental 0.8B-parameter typed-decision model built to run on a plain CPU using a custom Python runtime.

Bookmark Keep this in your collection.

Leave a noteWhat would you try with this?

Only you can see your notes.

Source screenshot of Compass by TDCC — SOURCE SCREENSHOTFull screenshot ↗

What it does

Compass fails its broader bilingual gate and scores below a lexical baseline on the English benchmark. This research experiment needs its custom Python runtime; standard Transformers, GGUF and Ollama loading are unsupported. Requests exceeding 2,048 tokens are rejected, not truncated.

How you can use it

The custom program runs on a plain processor, taking a situation and allowed answers and returning scores and decisions. An input that is too long is rejected rather than shortened. Keep its research status in mind: its English score trails word matching, and the broader bilingual checks fail.