| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.711935997009277, | |
| "base_token_generation_latency_sync": 39.71796798706055, | |
| "base_token_generation_latency_async": 39.712730795145035, | |
| "base_token_generation_throughput_sync": 0.025177521677991768, | |
| "base_token_generation_throughput_async": 0.025180842011555956, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 120.39700546264649, | |
| "base_inference_latency_async": 39.22069072723389, | |
| "base_inference_throughput_sync": 0.008305854420193638, | |
| "base_inference_throughput_async": 0.025496746269836203, | |
| "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": 9.126961708068848, | |
| "smashed_token_generation_latency_sync": 46.46282424926758, | |
| "smashed_token_generation_latency_async": 46.16003856062889, | |
| "smashed_token_generation_throughput_sync": 0.021522583186831645, | |
| "smashed_token_generation_throughput_async": 0.021663760065679976, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 164.03681182861328, | |
| "smashed_inference_latency_async": 70.84982395172119, | |
| "smashed_inference_throughput_sync": 0.006096192609771071, | |
| "smashed_inference_throughput_async": 0.014114361112335643, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |