| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 29282.24609375, | |
| "base_token_generation_latency_sync": 35.0454402923584, | |
| "base_token_generation_latency_async": 35.173794254660606, | |
| "base_token_generation_throughput_sync": 0.02853438255184508, | |
| "base_token_generation_throughput_async": 0.02843025670645406, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 126.72092132568359, | |
| "base_inference_latency_async": 31.963801383972168, | |
| "base_inference_throughput_sync": 0.007891356766811335, | |
| "base_inference_throughput_async": 0.03128539024464834, | |
| "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": 170638.125, | |
| "smashed_token_generation_latency_sync": 56.02240333557129, | |
| "smashed_token_generation_latency_async": 55.78384846448898, | |
| "smashed_token_generation_throughput_sync": 0.01785000179321212, | |
| "smashed_token_generation_throughput_async": 0.017926335803751196, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 212.82099151611328, | |
| "smashed_inference_latency_async": 109.1862678527832, | |
| "smashed_inference_throughput_sync": 0.004698784611781527, | |
| "smashed_inference_throughput_async": 0.009158660879849, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |