{ "arm": "coin", "seed": 314159, "n_mmlu": 40, "n_gsm8k": 40, "rows": [ { "condition": "no_aft", "capability": { "n": { "mmlu": 40, "gsm8k": 40 }, "mmlu": 0.675, "gsm8k": 0.75, "mean": 0.7125 }, "collapse": { "n": 80, "parseable_rate": 0.9875, "empty_rate": 0.0, "truncation_rate": 0.1875, "dispatch_intrusion_rate": 0.0, "repeated_fourgram_rate": 0.2625, "maximum_exact_response_share": 0.1, "response_chars": { "mean": 389.35, "median": 327.0, "max": 1274 } } }, { "condition": "step_4", "capability": { "n": { "mmlu": 40, "gsm8k": 40 }, "mmlu": 0.7, "gsm8k": 0.725, "mean": 0.7124999999999999 }, "collapse": { "n": 80, "parseable_rate": 0.9875, "empty_rate": 0.0, "truncation_rate": 0.2, "dispatch_intrusion_rate": 0.0, "repeated_fourgram_rate": 0.2, "maximum_exact_response_share": 0.1, "response_chars": { "mean": 384.975, "median": 357.5, "max": 1197 } } }, { "condition": "step_8", "capability": { "n": { "mmlu": 40, "gsm8k": 40 }, "mmlu": 0.725, "gsm8k": 0.75, "mean": 0.7375 }, "collapse": { "n": 80, "parseable_rate": 0.9875, "empty_rate": 0.0, "truncation_rate": 0.1375, "dispatch_intrusion_rate": 0.0, "repeated_fourgram_rate": 0.1375, "maximum_exact_response_share": 0.1125, "response_chars": { "mean": 333.125, "median": 290.5, "max": 1233 } } }, { "condition": "step_16", "capability": { "n": { "mmlu": 40, "gsm8k": 40 }, "mmlu": 0.725, "gsm8k": 0.9, "mean": 0.8125 }, "collapse": { "n": 80, "parseable_rate": 0.9875, "empty_rate": 0.0, "truncation_rate": 0.05, "dispatch_intrusion_rate": 0.0, "repeated_fourgram_rate": 0.1125, "maximum_exact_response_share": 0.1375, "response_chars": { "mean": 221.5625, "median": 141.5, "max": 951 } } }, { "condition": "step_32", "capability": { "n": { "mmlu": 40, "gsm8k": 40 }, "mmlu": 0.7, "gsm8k": 0.875, "mean": 0.7875 }, "collapse": { "n": 80, "parseable_rate": 1.0, "empty_rate": 0.0, "truncation_rate": 0.0375, "dispatch_intrusion_rate": 0.0, "repeated_fourgram_rate": 0.1, "maximum_exact_response_share": 0.15, "response_chars": { "mean": 199.7, "median": 63.0, "max": 951 } } }, { "condition": "step_64", "capability": { "n": { "mmlu": 40, "gsm8k": 40 }, "mmlu": 0.7, "gsm8k": 0.825, "mean": 0.7625 }, "collapse": { "n": 80, "parseable_rate": 1.0, "empty_rate": 0.0, "truncation_rate": 0.0375, "dispatch_intrusion_rate": 0.0, "repeated_fourgram_rate": 0.1125, "maximum_exact_response_share": 0.15, "response_chars": { "mean": 191.5375, "median": 31.5, "max": 995 } } }, { "condition": "step_128", "capability": { "n": { "mmlu": 40, "gsm8k": 40 }, "mmlu": 0.725, "gsm8k": 0.775, "mean": 0.75 }, "collapse": { "n": 80, "parseable_rate": 1.0, "empty_rate": 0.0, "truncation_rate": 0.0625, "dispatch_intrusion_rate": 0.0, "repeated_fourgram_rate": 0.1375, "maximum_exact_response_share": 0.1375, "response_chars": { "mean": 214.7125, "median": 63.5, "max": 878 } } }, { "condition": "step_256", "capability": { "n": { "mmlu": 40, "gsm8k": 40 }, "mmlu": 0.6, "gsm8k": 0.725, "mean": 0.6625 }, "collapse": { "n": 80, "parseable_rate": 0.9875, "empty_rate": 0.0, "truncation_rate": 0.0375, "dispatch_intrusion_rate": 0.0, "repeated_fourgram_rate": 0.125, "maximum_exact_response_share": 0.175, "response_chars": { "mean": 204.075, "median": 56.0, "max": 1006 } } }, { "condition": "step_512", "capability": { "n": { "mmlu": 40, "gsm8k": 40 }, "mmlu": 0.575, "gsm8k": 0.75, "mean": 0.6625 }, "collapse": { "n": 80, "parseable_rate": 1.0, "empty_rate": 0.0, "truncation_rate": 0.0375, "dispatch_intrusion_rate": 0.0, "repeated_fourgram_rate": 0.1, "maximum_exact_response_share": 0.175, "response_chars": { "mean": 203.8875, "median": 61.5, "max": 1094 } } }, { "condition": "step_1024", "capability": { "n": { "mmlu": 40, "gsm8k": 40 }, "mmlu": 0.6, "gsm8k": 0.65, "mean": 0.625 }, "collapse": { "n": 80, "parseable_rate": 1.0, "empty_rate": 0.0, "truncation_rate": 0.0375, "dispatch_intrusion_rate": 0.0, "repeated_fourgram_rate": 0.0875, "maximum_exact_response_share": 0.1375, "response_chars": { "mean": 184.025, "median": 53.5, "max": 1149 } } }, { "condition": "step_2048", "capability": { "n": { "mmlu": 40, "gsm8k": 40 }, "mmlu": 0.625, "gsm8k": 0.675, "mean": 0.65 }, "collapse": { "n": 80, "parseable_rate": 1.0, "empty_rate": 0.0, "truncation_rate": 0.05, "dispatch_intrusion_rate": 0.0, "repeated_fourgram_rate": 0.0875, "maximum_exact_response_share": 0.1375, "response_chars": { "mean": 188.1375, "median": 53.5, "max": 1140 } } } ] }