AniFileBERT / reports /perf_metrics.json
ModerRAS's picture
Train virtual-shard anime parser
359ff82
{
"sample_count": 1,
"samples": [
{
"step": 50.0,
"elapsed_seconds": 11.495871799997985,
"window_seconds": 11.495871799997985,
"steps_per_second": 4.349387403572886,
"samples_per_second": 7794.102227202612,
"tokens_per_second": 997645.0850819343,
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 122.5029296875,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 100.0,
"gpu_memory_util_percent": 75.0,
"gpu_memory_used_mb": 13744.27734375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 62.0,
"gpu_power_w": 217.742
}
],
"background_sample_count": 32,
"background_samples": [
{
"process_rss_mb": 2403.41796875,
"cuda_allocated_mb": 10657.99755859375,
"cuda_reserved_mb": 11200.0,
"cuda_max_allocated_mb": 11051.73876953125,
"cuda_max_reserved_mb": 11200.0,
"gpu_util_percent": 14.0,
"gpu_memory_util_percent": 5.0,
"gpu_memory_used_mb": 12818.02734375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 45.0,
"gpu_power_w": 42.684,
"elapsed_seconds": 0.541983800008893
},
{
"process_rss_mb": 2682.99609375,
"cuda_allocated_mb": 126.2216796875,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 23.0,
"gpu_memory_util_percent": 17.0,
"gpu_memory_used_mb": 13744.02734375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 49.0,
"gpu_power_w": 80.492,
"elapsed_seconds": 1.0560631999978796
},
{
"process_rss_mb": 2683.15625,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 84.0,
"gpu_memory_util_percent": 65.0,
"gpu_memory_used_mb": 13744.02734375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 59.0,
"gpu_power_w": 151.182,
"elapsed_seconds": 1.5636136999819428
},
{
"process_rss_mb": 2683.171875,
"cuda_allocated_mb": 126.2216796875,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 67.0,
"gpu_memory_util_percent": 48.0,
"gpu_memory_used_mb": 13744.02734375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 58.0,
"gpu_power_w": 198.581,
"elapsed_seconds": 2.090296700014733
},
{
"process_rss_mb": 2683.171875,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 100.0,
"gpu_memory_util_percent": 77.0,
"gpu_memory_used_mb": 13744.02734375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 58.0,
"gpu_power_w": 215.4,
"elapsed_seconds": 2.5990715000079945
},
{
"process_rss_mb": 2683.171875,
"cuda_allocated_mb": 126.2216796875,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 63.0,
"gpu_memory_util_percent": 47.0,
"gpu_memory_used_mb": 13744.02734375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 54.0,
"gpu_power_w": 209.578,
"elapsed_seconds": 3.1307488000020385
},
{
"process_rss_mb": 2683.171875,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 100.0,
"gpu_memory_util_percent": 74.0,
"gpu_memory_used_mb": 13744.02734375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 59.0,
"gpu_power_w": 215.475,
"elapsed_seconds": 3.644748599966988
},
{
"process_rss_mb": 2683.171875,
"cuda_allocated_mb": 126.2216796875,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 1.0,
"gpu_memory_util_percent": 0.0,
"gpu_memory_used_mb": 13744.02734375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 51.0,
"gpu_power_w": 199.707,
"elapsed_seconds": 4.402358899998944
},
{
"process_rss_mb": 2683.171875,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 100.0,
"gpu_memory_util_percent": 76.0,
"gpu_memory_used_mb": 13744.02734375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 60.0,
"gpu_power_w": 178.762,
"elapsed_seconds": 4.912791399983689
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 67.0,
"gpu_memory_util_percent": 51.0,
"gpu_memory_used_mb": 13744.02734375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 57.0,
"gpu_power_w": 187.267,
"elapsed_seconds": 5.420690300001297
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 82.0,
"gpu_memory_util_percent": 59.0,
"gpu_memory_used_mb": 13749.71484375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 61.0,
"gpu_power_w": 211.903,
"elapsed_seconds": 5.9282142000156455
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 100.0,
"gpu_memory_util_percent": 77.0,
"gpu_memory_used_mb": 13749.71484375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 60.0,
"gpu_power_w": 216.831,
"elapsed_seconds": 6.439060800010338
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 65.0,
"gpu_memory_util_percent": 47.0,
"gpu_memory_used_mb": 13749.71484375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 60.0,
"gpu_power_w": 214.227,
"elapsed_seconds": 6.946143499983009
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 100.0,
"gpu_memory_util_percent": 75.0,
"gpu_memory_used_mb": 13749.71484375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 62.0,
"gpu_power_w": 214.596,
"elapsed_seconds": 7.4573811999871396
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 5474.2958984375,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 64.0,
"gpu_memory_util_percent": 48.0,
"gpu_memory_used_mb": 13749.71484375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 56.0,
"gpu_power_w": 213.192,
"elapsed_seconds": 7.967849700013176
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 92.0,
"gpu_memory_util_percent": 67.0,
"gpu_memory_used_mb": 13745.71484375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 61.0,
"gpu_power_w": 217.827,
"elapsed_seconds": 8.474799200019334
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 1.0,
"gpu_memory_util_percent": 0.0,
"gpu_memory_used_mb": 13745.71484375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 51.0,
"gpu_power_w": 188.037,
"elapsed_seconds": 8.985988000000361
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 66.0,
"gpu_memory_util_percent": 49.0,
"gpu_memory_used_mb": 13744.52734375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 56.0,
"gpu_power_w": 171.267,
"elapsed_seconds": 9.494696600013413
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 93.0,
"gpu_memory_util_percent": 69.0,
"gpu_memory_used_mb": 13744.52734375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 62.0,
"gpu_power_w": 192.325,
"elapsed_seconds": 10.006947299989406
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 93.0,
"gpu_memory_util_percent": 71.0,
"gpu_memory_used_mb": 13744.52734375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 62.0,
"gpu_power_w": 221.52,
"elapsed_seconds": 10.517244499991648
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 65.0,
"gpu_memory_util_percent": 47.0,
"gpu_memory_used_mb": 13744.27734375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 61.0,
"gpu_power_w": 214.796,
"elapsed_seconds": 11.027270299964584
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 6567.63916015625,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 100.0,
"gpu_memory_util_percent": 75.0,
"gpu_memory_used_mb": 13744.27734375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 62.0,
"gpu_power_w": 217.742,
"elapsed_seconds": 11.53551159997005
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 64.0,
"gpu_memory_util_percent": 48.0,
"gpu_memory_used_mb": 13742.83984375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 57.0,
"gpu_power_w": 211.385,
"elapsed_seconds": 12.04551550000906
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 126.2216796875,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 99.0,
"gpu_memory_util_percent": 73.0,
"gpu_memory_used_mb": 13742.83984375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 63.0,
"gpu_power_w": 215.446,
"elapsed_seconds": 12.555750900006387
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 126.2216796875,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 86.0,
"gpu_memory_util_percent": 65.0,
"gpu_memory_used_mb": 13742.83984375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 64.0,
"gpu_power_w": 216.554,
"elapsed_seconds": 13.202394699968863
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 38.0,
"gpu_memory_util_percent": 27.0,
"gpu_memory_used_mb": 13742.83984375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 57.0,
"gpu_power_w": 182.772,
"elapsed_seconds": 13.7173714999808
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 126.2216796875,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 100.0,
"gpu_memory_util_percent": 75.0,
"gpu_memory_used_mb": 13742.83984375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 64.0,
"gpu_power_w": 178.689,
"elapsed_seconds": 14.244046800013166
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 66.0,
"gpu_memory_util_percent": 50.0,
"gpu_memory_used_mb": 13742.83984375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 59.0,
"gpu_power_w": 207.294,
"elapsed_seconds": 14.752780599985272
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 80.0,
"gpu_memory_util_percent": 58.0,
"gpu_memory_used_mb": 13742.83984375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 64.0,
"gpu_power_w": 214.991,
"elapsed_seconds": 15.258768500003498
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 66.0,
"gpu_memory_util_percent": 48.0,
"gpu_memory_used_mb": 13742.83984375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 61.0,
"gpu_power_w": 214.067,
"elapsed_seconds": 15.76770419999957
},
{
"process_rss_mb": 2683.17578125,
"cuda_allocated_mb": 141.64453125,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 100.0,
"gpu_memory_util_percent": 75.0,
"gpu_memory_used_mb": 13742.83984375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 65.0,
"gpu_power_w": 218.24,
"elapsed_seconds": 16.273408600012772
},
{
"process_rss_mb": 2720.89453125,
"cuda_allocated_mb": 139.14794921875,
"cuda_reserved_mb": 12124.0,
"cuda_max_allocated_mb": 11120.271484375,
"cuda_max_reserved_mb": 12124.0,
"gpu_util_percent": 93.0,
"gpu_memory_util_percent": 70.0,
"gpu_memory_used_mb": 13742.96484375,
"gpu_memory_total_mb": 16303.0,
"gpu_temperature_c": 62.0,
"gpu_power_w": 221.991,
"elapsed_seconds": 16.781973899982404
}
],
"step_samples_per_second_avg": 7794.102227202612,
"step_samples_per_second_max": 7794.102227202612,
"step_samples_per_second_min": 7794.102227202612,
"step_tokens_per_second_avg": 997645.0850819343,
"step_tokens_per_second_max": 997645.0850819343,
"step_tokens_per_second_min": 997645.0850819343,
"step_process_rss_mb_avg": 2683.17578125,
"step_process_rss_mb_max": 2683.17578125,
"step_process_rss_mb_min": 2683.17578125,
"step_cuda_max_allocated_mb_avg": 11120.271484375,
"step_cuda_max_allocated_mb_max": 11120.271484375,
"step_cuda_max_allocated_mb_min": 11120.271484375,
"step_gpu_util_percent_avg": 100.0,
"step_gpu_util_percent_max": 100.0,
"step_gpu_util_percent_min": 100.0,
"step_gpu_memory_util_percent_avg": 75.0,
"step_gpu_memory_util_percent_max": 75.0,
"step_gpu_memory_util_percent_min": 75.0,
"step_gpu_power_w_avg": 217.742,
"step_gpu_power_w_max": 217.742,
"step_gpu_power_w_min": 217.742,
"step_gpu_temperature_c_avg": 62.0,
"step_gpu_temperature_c_max": 62.0,
"step_gpu_temperature_c_min": 62.0,
"background_process_rss_mb_avg": 2675.6051025390625,
"background_process_rss_mb_max": 2720.89453125,
"background_process_rss_mb_min": 2403.41796875,
"background_cuda_max_allocated_mb_avg": 11118.129837036133,
"background_cuda_max_allocated_mb_max": 11120.271484375,
"background_cuda_max_allocated_mb_min": 11051.73876953125,
"background_gpu_util_percent_avg": 72.875,
"background_gpu_util_percent_max": 100.0,
"background_gpu_util_percent_min": 1.0,
"background_gpu_memory_util_percent_avg": 54.15625,
"background_gpu_memory_util_percent_max": 77.0,
"background_gpu_memory_util_percent_min": 0.0,
"background_gpu_power_w_avg": 195.463125,
"background_gpu_power_w_max": 221.991,
"background_gpu_power_w_min": 42.684,
"background_gpu_temperature_c_avg": 58.75,
"background_gpu_temperature_c_max": 65.0,
"background_gpu_temperature_c_min": 45.0,
"samples_per_second_avg": 7794.102227202612,
"samples_per_second_max": 7794.102227202612,
"tokens_per_second_avg": 997645.0850819343,
"tokens_per_second_max": 997645.0850819343,
"process_rss_mb_avg": 2683.17578125,
"process_rss_mb_max": 2683.17578125,
"cuda_max_allocated_mb_avg": 11120.271484375,
"cuda_max_allocated_mb_max": 11120.271484375,
"gpu_util_percent_avg": 100.0,
"gpu_util_percent_max": 100.0,
"gpu_memory_util_percent_avg": 75.0,
"gpu_memory_util_percent_max": 75.0,
"gpu_power_w_avg": 217.742,
"gpu_power_w_max": 217.742,
"gpu_temperature_c_avg": 62.0,
"gpu_temperature_c_max": 62.0
}