| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.711685180664062, | |
| "base_token_generation_latency_sync": 41.17222557067871, | |
| "base_token_generation_latency_async": 39.11548051983118, | |
| "base_token_generation_throughput_sync": 0.024288218237883206, | |
| "base_token_generation_throughput_async": 0.025565325715301117, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.87609558105468, | |
| "base_inference_latency_async": 39.180731773376465, | |
| "base_inference_throughput_sync": 0.008341946700489974, | |
| "base_inference_throughput_async": 0.025522749441844416, | |
| "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": 216470.109375, | |
| "smashed_token_generation_latency_sync": 175.36331787109376, | |
| "smashed_token_generation_latency_async": 175.5636254325509, | |
| "smashed_token_generation_throughput_sync": 0.005702446852283447, | |
| "smashed_token_generation_throughput_async": 0.005695940702615452, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 271.59930725097655, | |
| "smashed_inference_latency_async": 233.75272750854492, | |
| "smashed_inference_throughput_sync": 0.0036818945163064453, | |
| "smashed_inference_throughput_async": 0.004278024948236998, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |