| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 7.200127124786377, | |
| "base_token_generation_latency_sync": 50.63789672851563, | |
| "base_token_generation_latency_async": 49.18233975768089, | |
| "base_token_generation_throughput_sync": 0.019748055598779873, | |
| "base_token_generation_throughput_async": 0.020332501563100774, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 104.84735946655273, | |
| "base_inference_latency_async": 101.91307067871094, | |
| "base_inference_throughput_sync": 0.009537674626121692, | |
| "base_inference_throughput_async": 0.009812284070534775, | |
| "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": 8.361743927001953, | |
| "smashed_token_generation_latency_sync": 66.61556053161621, | |
| "smashed_token_generation_latency_async": 64.81923274695873, | |
| "smashed_token_generation_throughput_sync": 0.015011507702099016, | |
| "smashed_token_generation_throughput_async": 0.015427519852075374, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 146.04206237792968, | |
| "smashed_inference_latency_async": 136.93158626556396, | |
| "smashed_inference_throughput_sync": 0.006847342359574367, | |
| "smashed_inference_throughput_async": 0.007302916932990234, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |