| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 10.639251708984375, | |
| "base_token_generation_latency_sync": 40.07466354370117, | |
| "base_token_generation_latency_async": 40.20857363939285, | |
| "base_token_generation_throughput_sync": 0.024953422226727025, | |
| "base_token_generation_throughput_async": 0.02487031768319897, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.12806396484375, | |
| "base_inference_latency_async": 40.34836292266846, | |
| "base_inference_throughput_sync": 0.008394327639666108, | |
| "base_inference_throughput_async": 0.024784153000621038, | |
| "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": 11.474609375, | |
| "smashed_token_generation_latency_sync": 47.03260345458985, | |
| "smashed_token_generation_latency_async": 46.74163851886988, | |
| "smashed_token_generation_throughput_sync": 0.02126184660318674, | |
| "smashed_token_generation_throughput_async": 0.021394200795855585, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 162.76981506347656, | |
| "smashed_inference_latency_async": 70.5676794052124, | |
| "smashed_inference_throughput_sync": 0.006143645242885006, | |
| "smashed_inference_throughput_async": 0.014170793321087106, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |