nice result! I've been looking into benchmarking models recently, it would be interesting to run your model through on the same battery of tests [https://github.com/emrgnt-cmplxty/zero-shot-replication/blob...]