| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 9.401168823242188, | |
| "base_token_generation_latency_sync": 38.77763404846191, | |
| "base_token_generation_latency_async": 38.67101036012173, | |
| "base_token_generation_throughput_sync": 0.02578806119915055, | |
| "base_token_generation_throughput_async": 0.025859164027201596, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 121.23289642333984, | |
| "base_inference_latency_async": 44.76165771484375, | |
| "base_inference_throughput_sync": 0.008248586229500324, | |
| "base_inference_throughput_async": 0.022340548832452702, | |
| "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": 9.998388290405273, | |
| "smashed_token_generation_latency_sync": 62.574696731567386, | |
| "smashed_token_generation_latency_async": 62.1792558580637, | |
| "smashed_token_generation_throughput_sync": 0.01598090046348598, | |
| "smashed_token_generation_throughput_async": 0.016082534057382344, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 195.8287338256836, | |
| "smashed_inference_latency_async": 110.43708324432373, | |
| "smashed_inference_throughput_sync": 0.005106502914378986, | |
| "smashed_inference_throughput_async": 0.0090549294731704, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |