HaolinRPI/

Trace

MIA experiment splits with model reasoning trace prepended to output.

output = raw_text + ground_truth input = model question / prompt

	Structure




	

Subset train test val

chatdoctor 18,000 2,000 200

codealpaca 18,000 2,000 200

wealth 18,000 2,000 200

Each file is JSONL with fields: input, output.