| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 6.637787342071533, | |
| "base_token_generation_latency_sync": 39.10443000793457, | |
| "base_token_generation_latency_async": 35.745227709412575, | |
| "base_token_generation_throughput_sync": 0.02557255021482458, | |
| "base_token_generation_throughput_async": 0.027975762474627516, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.40597763061524, | |
| "base_inference_latency_async": 38.70058059692383, | |
| "base_inference_throughput_sync": 0.008374790105513142, | |
| "base_inference_throughput_async": 0.025839405625854784, | |
| "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": 7.146455764770508, | |
| "smashed_token_generation_latency_sync": 42.05969581604004, | |
| "smashed_token_generation_latency_async": 42.0763123780489, | |
| "smashed_token_generation_throughput_sync": 0.023775730675128573, | |
| "smashed_token_generation_throughput_async": 0.023766341285214373, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 163.19764251708983, | |
| "smashed_inference_latency_async": 70.17691135406494, | |
| "smashed_inference_throughput_sync": 0.006127539494912013, | |
| "smashed_inference_throughput_async": 0.014249700944441406, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |