| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 10.639251708984375, | |
| "base_token_generation_latency_sync": 36.38071460723877, | |
| "base_token_generation_latency_async": 37.1865913271904, | |
| "base_token_generation_throughput_sync": 0.027487090641177984, | |
| "base_token_generation_throughput_async": 0.026891413391493393, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 117.6410125732422, | |
| "base_inference_latency_async": 40.06998538970947, | |
| "base_inference_throughput_sync": 0.00850043686403506, | |
| "base_inference_throughput_async": 0.024956335528308275, | |
| "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": 12.093978881835938, | |
| "smashed_token_generation_latency_sync": 60.57174797058106, | |
| "smashed_token_generation_latency_async": 61.36605739593506, | |
| "smashed_token_generation_throughput_sync": 0.016509346906839927, | |
| "smashed_token_generation_throughput_async": 0.016295653369874808, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 192.84817657470703, | |
| "smashed_inference_latency_async": 110.32328605651855, | |
| "smashed_inference_throughput_sync": 0.005185426265166744, | |
| "smashed_inference_throughput_async": 0.009064269527719656, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |