| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 9.883513450622559, | |
| "base_token_generation_latency_sync": 32.2106897354126, | |
| "base_token_generation_latency_async": 32.60008953511715, | |
| "base_token_generation_throughput_sync": 0.03104559412462984, | |
| "base_token_generation_throughput_async": 0.03067476237826862, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.63412399291992, | |
| "base_inference_latency_async": 38.892364501953125, | |
| "base_inference_throughput_sync": 0.008358819094618698, | |
| "base_inference_throughput_async": 0.025711987759185515, | |
| "base_inference_CO2_emissions": null, | |
| "base_inference_energy_consumption": null, | |
| "smashed_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "smashed_current_gpu_total_memory": 40339.3125, | |
| "smashed_perplexity": 222301.21875, | |
| "smashed_token_generation_latency_sync": 167.92325897216796, | |
| "smashed_token_generation_latency_async": 167.88078993558884, | |
| "smashed_token_generation_throughput_sync": 0.005955101193967076, | |
| "smashed_token_generation_throughput_async": 0.005956607664186426, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 263.06273193359374, | |
| "smashed_inference_latency_async": 210.80782413482666, | |
| "smashed_inference_throughput_sync": 0.0038013746479772553, | |
| "smashed_inference_throughput_async": 0.0047436569496606, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |