| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 47.838710403442384, | |
| "base_token_generation_latency_async": 47.88779690861702, | |
| "base_token_generation_throughput_sync": 0.02090357351957468, | |
| "base_token_generation_throughput_async": 0.020882146696125377, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 46.498303985595705, | |
| "base_inference_latency_async": 45.331716537475586, | |
| "base_inference_throughput_sync": 0.021506160747492665, | |
| "base_inference_throughput_async": 0.022059610276908514, | |
| "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_token_generation_latency_sync": 188.36228790283204, | |
| "smashed_token_generation_latency_async": 188.33214957267046, | |
| "smashed_token_generation_throughput_sync": 0.005308918314455051, | |
| "smashed_token_generation_throughput_async": 0.005309767887580643, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 194.22156677246093, | |
| "smashed_inference_latency_async": 166.04604721069336, | |
| "smashed_inference_throughput_sync": 0.005148758794493424, | |
| "smashed_inference_throughput_async": 0.00602242580777075, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |