AdamRoch/qwen3-0.6b-nfl-play-normalizer-v2-n-12500

NFL play normalizer v2, N/2

This public repository contains the locked N/2 QLoRA adapter from the SLM-8 data-efficiency curve. It trained for one epoch on 12500 unchanged real 2019 through 2022 nflverse records. Load it onto Qwen/Qwen3-0.6B at revision c1899de289a04d12100db370d81485cdf75e47ca.

On the frozen 30-record 2023 discovery set it produced 26/30 exact records and 30/30 valid JSON responses. These are discovery results, not final-set or production claims. Runtime configuration, raw responses, predictions, scores, and checksums are retained under metadata/.

Code revision: 1a6632c5b53d5d001d808de6bb353af785cd1480. Training data derives from CC BY 4.0 nflverse releases and requires attribution.