| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.711685180664062, | |
| "base_token_generation_latency_sync": 35.84392547607422, | |
| "base_token_generation_latency_async": 36.515641771256924, | |
| "base_token_generation_throughput_sync": 0.02789873002797919, | |
| "base_token_generation_throughput_async": 0.027385524435370713, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.2900604248047, | |
| "base_inference_latency_async": 38.611435890197754, | |
| "base_inference_throughput_sync": 0.008382928103472267, | |
| "base_inference_throughput_async": 0.02589906272441603, | |
| "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": 9.146251678466797, | |
| "smashed_token_generation_latency_sync": 42.331953811645505, | |
| "smashed_token_generation_latency_async": 42.045436799526215, | |
| "smashed_token_generation_throughput_sync": 0.023622817043821406, | |
| "smashed_token_generation_throughput_async": 0.023783793822098394, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 162.66905670166017, | |
| "smashed_inference_latency_async": 69.82095241546631, | |
| "smashed_inference_throughput_sync": 0.006147450660109436, | |
| "smashed_inference_throughput_async": 0.0143223483124313, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |