| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 9.883513450622559, | |
| "base_token_generation_latency_sync": 38.06065979003906, | |
| "base_token_generation_latency_async": 38.11613507568836, | |
| "base_token_generation_throughput_sync": 0.02627384826002707, | |
| "base_token_generation_throughput_async": 0.02623560856876674, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.0922233581543, | |
| "base_inference_latency_async": 38.748931884765625, | |
| "base_inference_throughput_sync": 0.008396853898617971, | |
| "base_inference_throughput_async": 0.02580716296835929, | |
| "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": 10.680657386779785, | |
| "smashed_token_generation_latency_sync": 167.25687408447266, | |
| "smashed_token_generation_latency_async": 168.01605839282274, | |
| "smashed_token_generation_throughput_sync": 0.005978827509923165, | |
| "smashed_token_generation_throughput_async": 0.005951812044429663, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 264.7478271484375, | |
| "smashed_inference_latency_async": 196.02723121643066, | |
| "smashed_inference_throughput_sync": 0.0037771792530683357, | |
| "smashed_inference_throughput_async": 0.005101332063890222, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |