| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 6.637787342071533, | |
| "base_token_generation_latency_sync": 37.99737701416016, | |
| "base_token_generation_latency_async": 37.817021273076534, | |
| "base_token_generation_throughput_sync": 0.026317606071264828, | |
| "base_token_generation_throughput_async": 0.02644311916528287, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.08403244018555, | |
| "base_inference_latency_async": 38.645267486572266, | |
| "base_inference_throughput_sync": 0.008397431456667272, | |
| "base_inference_throughput_async": 0.025876389660065394, | |
| "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": 152463.125, | |
| "smashed_token_generation_latency_sync": 176.9831344604492, | |
| "smashed_token_generation_latency_async": 178.20135485380888, | |
| "smashed_token_generation_throughput_sync": 0.005650255901776178, | |
| "smashed_token_generation_throughput_async": 0.005611629613143909, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 271.66566314697263, | |
| "smashed_inference_latency_async": 233.64920616149902, | |
| "smashed_inference_throughput_sync": 0.0036809951924582913, | |
| "smashed_inference_throughput_async": 0.004279920383332255, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |