Model card: 72-28 result, what produced the 12-point gain, limitations

Add LoRA adapter weights (Llama-3.3-70B, math and code)

initial commit