| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.629734992980957, | |
| "base_token_generation_latency_sync": 35.77475910186767, | |
| "base_token_generation_latency_async": 36.34010720998049, | |
| "base_token_generation_throughput_sync": 0.02795266900756834, | |
| "base_token_generation_throughput_async": 0.027517805443495195, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 117.67265243530274, | |
| "base_inference_latency_async": 38.82002830505371, | |
| "base_inference_throughput_sync": 0.008498151263734002, | |
| "base_inference_throughput_async": 0.025759898785798076, | |
| "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": 9.385024070739746, | |
| "smashed_token_generation_latency_sync": 59.95228500366211, | |
| "smashed_token_generation_latency_async": 60.92198956757784, | |
| "smashed_token_generation_throughput_sync": 0.016679931381079406, | |
| "smashed_token_generation_throughput_async": 0.016414434379079954, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 192.92713012695313, | |
| "smashed_inference_latency_async": 108.5883617401123, | |
| "smashed_inference_throughput_sync": 0.005183304179883687, | |
| "smashed_inference_throughput_async": 0.009209090034835677, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |