| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 10.554253578186035, | |
| "base_token_generation_latency_sync": 37.02315559387207, | |
| "base_token_generation_latency_async": 36.12542040646076, | |
| "base_token_generation_throughput_sync": 0.027010123366294474, | |
| "base_token_generation_throughput_async": 0.02768133875671541, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 118.97077789306641, | |
| "base_inference_latency_async": 38.97838592529297, | |
| "base_inference_throughput_sync": 0.00840542541378373, | |
| "base_inference_throughput_async": 0.02565524395793677, | |
| "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": 317970.84375, | |
| "smashed_token_generation_latency_sync": 176.74646911621093, | |
| "smashed_token_generation_latency_async": 176.82972494512796, | |
| "smashed_token_generation_throughput_sync": 0.005657821652677538, | |
| "smashed_token_generation_throughput_async": 0.005655157809640376, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 270.4947204589844, | |
| "smashed_inference_latency_async": 233.14297199249268, | |
| "smashed_inference_throughput_sync": 0.0036969298265902084, | |
| "smashed_inference_throughput_async": 0.004289213573344173, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |