| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.633415222167969, | |
| "base_token_generation_latency_sync": 34.05241985321045, | |
| "base_token_generation_latency_async": 35.20148601382971, | |
| "base_token_generation_throughput_sync": 0.0293664886169821, | |
| "base_token_generation_throughput_async": 0.028407891632959106, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.67354888916016, | |
| "base_inference_latency_async": 43.74585151672363, | |
| "base_inference_throughput_sync": 0.008356065390240787, | |
| "base_inference_throughput_async": 0.022859310433532405, | |
| "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": 86191.7109375, | |
| "smashed_token_generation_latency_sync": 166.4123962402344, | |
| "smashed_token_generation_latency_async": 166.60868488252163, | |
| "smashed_token_generation_throughput_sync": 0.006009167721834805, | |
| "smashed_token_generation_throughput_async": 0.006002088070649591, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 262.3994888305664, | |
| "smashed_inference_latency_async": 209.65845584869385, | |
| "smashed_inference_throughput_sync": 0.00381098303375777, | |
| "smashed_inference_throughput_async": 0.004769662143852091, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |