drproduck/

r1-qwen14b-awq-matharena-n32

using casperhansen/deepseek-r1-distill-qwen-14b-awq to generate 32 samples for each problem in matharena. Note that the prompt was appended with after modulo 1000 (a requirement from aimo2), leading to several incorrect answers.

• name: problem dtype: string
• name: answer dtype: string
• name: completion sequence: string
• name: prediction sequence: string
• name: voted_prediction dtype: int64 splits:
• name: train num_bytes:… See the full description on the dataset page: https://huggingface.co/datasets/drproduck/r1-qwen14b-awq-matharena-n32.