| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 6.637787342071533, | |
| "base_token_generation_latency_sync": 41.50432357788086, | |
| "base_token_generation_latency_async": 42.54915565252304, | |
| "base_token_generation_throughput_sync": 0.024093875379598666, | |
| "base_token_generation_throughput_async": 0.023502229002297557, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.23496932983399, | |
| "base_inference_latency_async": 45.11885643005371, | |
| "base_inference_throughput_sync": 0.008386801335384655, | |
| "base_inference_throughput_async": 0.022163682307645968, | |
| "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": 7.51709508895874, | |
| "smashed_token_generation_latency_sync": 73.48166198730469, | |
| "smashed_token_generation_latency_async": 74.85278900712729, | |
| "smashed_token_generation_throughput_sync": 0.013608837537898481, | |
| "smashed_token_generation_throughput_async": 0.013359555646013172, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 194.81293029785155, | |
| "smashed_inference_latency_async": 110.41016578674316, | |
| "smashed_inference_throughput_sync": 0.0051331295026007225, | |
| "smashed_inference_throughput_async": 0.009057137020620876, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |