| { | |
| "current_gpu_type": "Tesla T4", | |
| "current_gpu_total_memory": 15095.0625, | |
| "perplexity": 2.2858734130859375, | |
| "memory_inference_first": 1178.0, | |
| "memory_inference": 1194.0, | |
| "token_generation_latency_sync": 265.68286819458007, | |
| "token_generation_latency_async": 265.86400624364614, | |
| "token_generation_throughput_sync": 0.0037638858944703305, | |
| "token_generation_throughput_async": 0.0037613214896173967, | |
| "token_generation_CO2_emissions": 8.567858013593123e-05, | |
| "token_generation_energy_consumption": 0.01875942725794068, | |
| "inference_latency_sync": 256.0767112731934, | |
| "inference_latency_async": 41.45829677581787, | |
| "inference_throughput_sync": 0.003905079829509206, | |
| "inference_throughput_async": 0.024120624284384207, | |
| "inference_CO2_emissions": 8.569058627578585e-05, | |
| "inference_energy_consumption": 0.00029453419207313414 | |
| } |