| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 53.76724815368652, | |
| "base_token_generation_latency_async": 53.66944447159767, | |
| "base_token_generation_throughput_sync": 0.018598682922020354, | |
| "base_token_generation_throughput_async": 0.018632575944198725, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 53.00592613220215, | |
| "base_inference_latency_async": 52.15179920196533, | |
| "base_inference_throughput_sync": 0.01886581506954333, | |
| "base_inference_throughput_async": 0.019174793876762648, | |
| "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": 163.97110137939453, | |
| "smashed_token_generation_latency_async": 164.70238156616688, | |
| "smashed_token_generation_throughput_sync": 0.006098635622909009, | |
| "smashed_token_generation_throughput_async": 0.006071557621030902, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 171.56013946533204, | |
| "smashed_inference_latency_async": 128.83808612823486, | |
| "smashed_inference_throughput_sync": 0.005828859798765055, | |
| "smashed_inference_throughput_async": 0.007761680028409317, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |