| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 76.9214370727539, | |
| "base_token_generation_latency_async": 76.84411387890577, | |
| "base_token_generation_throughput_sync": 0.01300027714061269, | |
| "base_token_generation_throughput_async": 0.01301335846719298, | |
| "base_token_generation_CO2_emissions": 4.107156647942229e-05, | |
| "base_token_generation_energy_consumption": 0.013795784438629225, | |
| "base_inference_latency_sync": 75.23112869262695, | |
| "base_inference_latency_async": 74.10402297973633, | |
| "base_inference_throughput_sync": 0.01329237002525532, | |
| "base_inference_throughput_async": 0.013494544017852432, | |
| "base_inference_CO2_emissions": 4.120221930394127e-05, | |
| "base_inference_energy_consumption": 0.00011383432613620926, | |
| "smashed_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "smashed_current_gpu_total_memory": 40339.3125, | |
| "smashed_token_generation_latency_sync": 49.12493553161621, | |
| "smashed_token_generation_latency_async": 48.867019824683666, | |
| "smashed_token_generation_throughput_sync": 0.02035626081089536, | |
| "smashed_token_generation_throughput_async": 0.020463699312698435, | |
| "smashed_token_generation_CO2_emissions": 3.728837750948082e-05, | |
| "smashed_token_generation_energy_consumption": 0.009091379798953737, | |
| "smashed_inference_latency_sync": 52.04664306640625, | |
| "smashed_inference_latency_async": 49.13980960845947, | |
| "smashed_inference_throughput_sync": 0.01921353503479756, | |
| "smashed_inference_throughput_async": 0.02035009919590427, | |
| "smashed_inference_CO2_emissions": 3.594376337961103e-05, | |
| "smashed_inference_energy_consumption": 0.00010089579813269325 | |
| } |