architecture.py and training.py). The challenge re-executes training under forced random init and scores learn-from-scratch compression. Full miner pack:
PRISM miner documentation
Overview, how it works, quickstart, submit, scoring, constraints, examples, API, and challenge operators.
One-line flow
- Submit two-script bundle through the PRISM challenge routes
- Static AST sandbox and LLM hard gate
- GPU re-execution on locked FineWeb-Edu train split
- Challenge-computed prequential bits-per-byte score
- Raw weights via
get_weights; BASE normalizes for netuid 100