| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 10.554253578186035, | |
| "base_token_generation_latency_sync": 39.642577743530275, | |
| "base_token_generation_latency_async": 40.50944782793522, | |
| "base_token_generation_throughput_sync": 0.02522540301161928, | |
| "base_token_generation_throughput_async": 0.024685599375422796, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.96098556518555, | |
| "base_inference_latency_async": 39.08228874206543, | |
| "base_inference_throughput_sync": 0.008336043550230841, | |
| "base_inference_throughput_async": 0.02558703781653581, | |
| "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": 11.702524185180664, | |
| "smashed_token_generation_latency_sync": 166.81119079589843, | |
| "smashed_token_generation_latency_async": 166.94455239921808, | |
| "smashed_token_generation_throughput_sync": 0.005994801639079169, | |
| "smashed_token_generation_throughput_async": 0.0059900127654880205, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 265.7383438110352, | |
| "smashed_inference_latency_async": 196.71142101287842, | |
| "smashed_inference_throughput_sync": 0.003763100144520708, | |
| "smashed_inference_throughput_async": 0.005083588918482427, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |