| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 10.554253578186035, | |
| "base_token_generation_latency_sync": 39.53613395690918, | |
| "base_token_generation_latency_async": 39.30522073060274, | |
| "base_token_generation_throughput_sync": 0.02529331778089152, | |
| "base_token_generation_throughput_async": 0.025441912840382747, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.83032455444337, | |
| "base_inference_latency_async": 40.5956506729126, | |
| "base_inference_throughput_sync": 0.00834513303471579, | |
| "base_inference_throughput_async": 0.024633180733010122, | |
| "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": 11.264032363891602, | |
| "smashed_token_generation_latency_sync": 46.946524429321286, | |
| "smashed_token_generation_latency_async": 45.606798492372036, | |
| "smashed_token_generation_throughput_sync": 0.02130083136410908, | |
| "smashed_token_generation_throughput_async": 0.02192655553682978, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 164.13562622070313, | |
| "smashed_inference_latency_async": 71.22159004211426, | |
| "smashed_inference_throughput_sync": 0.006092522525581139, | |
| "smashed_inference_throughput_async": 0.01404068624989539, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |