| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 11.945297241210938, | |
| "base_token_generation_latency_sync": 39.27951927185059, | |
| "base_token_generation_latency_async": 38.697187788784504, | |
| "base_token_generation_throughput_sync": 0.02545856004700759, | |
| "base_token_generation_throughput_async": 0.025841671117244006, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 120.78284759521485, | |
| "base_inference_latency_async": 38.20943832397461, | |
| "base_inference_throughput_sync": 0.008279321277068631, | |
| "base_inference_throughput_async": 0.02617154409654191, | |
| "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": 131872.640625, | |
| "smashed_token_generation_latency_sync": 176.11474609375, | |
| "smashed_token_generation_latency_async": 176.4693534001708, | |
| "smashed_token_generation_throughput_sync": 0.005678116240576905, | |
| "smashed_token_generation_throughput_async": 0.005666706318871978, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 273.20965270996095, | |
| "smashed_inference_latency_async": 232.32975006103516, | |
| "smashed_inference_throughput_sync": 0.0036601927863127107, | |
| "smashed_inference_throughput_async": 0.004304227072672745, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |