| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.891793251037598, | |
| "base_token_generation_latency_sync": 33.70677719116211, | |
| "base_token_generation_latency_async": 33.44351705163717, | |
| "base_token_generation_throughput_sync": 0.029667624238552216, | |
| "base_token_generation_throughput_async": 0.02990116136577348, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 120.11499710083008, | |
| "base_inference_latency_async": 38.821959495544434, | |
| "base_inference_throughput_sync": 0.008325355069197177, | |
| "base_inference_throughput_async": 0.025758617364864574, | |
| "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": 285619.03125, | |
| "smashed_token_generation_latency_sync": 174.61234588623046, | |
| "smashed_token_generation_latency_async": 174.5924288406968, | |
| "smashed_token_generation_throughput_sync": 0.005726971909830218, | |
| "smashed_token_generation_throughput_async": 0.005727625227737848, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 267.6000747680664, | |
| "smashed_inference_latency_async": 230.53014278411865, | |
| "smashed_inference_throughput_sync": 0.0037369197331754006, | |
| "smashed_inference_throughput_async": 0.004337827530590896, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |