| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 6.637787342071533, | |
| "base_token_generation_latency_sync": 40.13633842468262, | |
| "base_token_generation_latency_async": 38.44778724014759, | |
| "base_token_generation_throughput_sync": 0.02491507793807695, | |
| "base_token_generation_throughput_async": 0.026009299150401802, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 117.65217208862305, | |
| "base_inference_latency_async": 38.692593574523926, | |
| "base_inference_throughput_sync": 0.00849963058265288, | |
| "base_inference_throughput_async": 0.02584473946089834, | |
| "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": 20686.58203125, | |
| "smashed_token_generation_latency_sync": 168.5132278442383, | |
| "smashed_token_generation_latency_async": 168.39814521372318, | |
| "smashed_token_generation_throughput_sync": 0.0059342522411613245, | |
| "smashed_token_generation_throughput_async": 0.005938307685817121, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 262.7811309814453, | |
| "smashed_inference_latency_async": 210.38615703582764, | |
| "smashed_inference_throughput_sync": 0.0038054482689269227, | |
| "smashed_inference_throughput_async": 0.004753164438617058, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |