RLAIF (DPO (Direct Preference Optimization)) refinement of sudhisrk1982/slm-125m-instruct, trained on 500 AI-generated preference triplets (chosen from Gemini 2.5 Flash-Lite, rejected sampled from the SFT model itself, judged by Gemini 2.5 Flash).

Result: coherent improvement. Answers are more complete and on-topic than the SFT baseline. DPO was the most stable of the three methods (val loss 0.30, preference accuracy 1.00).

Not legal advice.