| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 10.639251708984375, | |
| "base_token_generation_latency_sync": 37.90767402648926, | |
| "base_token_generation_latency_async": 36.66847534477711, | |
| "base_token_generation_throughput_sync": 0.026379882851720642, | |
| "base_token_generation_throughput_async": 0.027271382041316194, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 120.66785278320313, | |
| "base_inference_latency_async": 40.3106689453125, | |
| "base_inference_throughput_sync": 0.008287211356919075, | |
| "base_inference_throughput_async": 0.0248073283367401, | |
| "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": 124304.8828125, | |
| "smashed_token_generation_latency_sync": 169.84294891357422, | |
| "smashed_token_generation_latency_async": 169.60067860782146, | |
| "smashed_token_generation_throughput_sync": 0.005887792259829739, | |
| "smashed_token_generation_throughput_async": 0.005896202823058062, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 265.8951187133789, | |
| "smashed_inference_latency_async": 211.9481086730957, | |
| "smashed_inference_throughput_sync": 0.0037608813762314603, | |
| "smashed_inference_throughput_async": 0.004718135992156358, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |