model_path: joesharratt29/mini_sudoku_non_curriculum # Default model path # model_path: joesharratt29/mini_sudoku_non_curriculum # model_path: joesharratt29/mini_sudoku_curriculum max_tokens: 2048 # From max_response_length in training config top_p: 1.0 temperature: 1.0 # developer_prompt: DeepSeekZero developer_role: system output_dir: results save_metadata: true save_full_results: true eval_repeats: 3 categories: - category: algorithmic datasets: - dataset: mini_sudoku size: 100 seed: 42 params: min_empty: 4 max_empty: 12