| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.405904769897461, | |
| "base_token_generation_latency_sync": 35.634182929992676, | |
| "base_token_generation_latency_async": 36.07999123632908, | |
| "base_token_generation_throughput_sync": 0.02806294175355757, | |
| "base_token_generation_throughput_async": 0.027716192984911157, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 120.22138977050781, | |
| "base_inference_latency_async": 39.18566703796387, | |
| "base_inference_throughput_sync": 0.008317987355735225, | |
| "base_inference_throughput_async": 0.02551953496239275, | |
| "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": 8.853601455688477, | |
| "smashed_token_generation_latency_sync": 40.49961433410645, | |
| "smashed_token_generation_latency_async": 40.31131472438574, | |
| "smashed_token_generation_throughput_sync": 0.02469159315321819, | |
| "smashed_token_generation_throughput_async": 0.02480693092837939, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 163.6304901123047, | |
| "smashed_inference_latency_async": 70.32005786895752, | |
| "smashed_inference_throughput_sync": 0.006111330469728893, | |
| "smashed_inference_throughput_async": 0.014220693644244647, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |