loan-default-explainer-1b-v2.0

LoRA fine-tune of unsloth/Llama-3.2-1B-Instruct, trained to write structured markdown loan assessment memos from SHAP-based credit risk model output.

• Format: fp16 merged safetensors (not quantized)
• Base model: unsloth/Llama-3.2-1B-Instruct
• Chat template: standard Llama-3.2 template (system/user/assistant), unmodified — including the default "Cutting Knowledge Date / Today Date" system-turn boilerplate, since training data was generated with it present. Serve with the same template; stripping it at inference time will not match training.