| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 7.280252456665039, | |
| "base_token_generation_latency_sync": 33.7761251449585, | |
| "base_token_generation_latency_async": 33.97261742502451, | |
| "base_token_generation_throughput_sync": 0.029606711714510046, | |
| "base_token_generation_throughput_async": 0.02943547114692999, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 124.69340133666992, | |
| "base_inference_latency_async": 45.56310176849365, | |
| "base_inference_throughput_sync": 0.00801967056219774, | |
| "base_inference_throughput_async": 0.02194758392615597, | |
| "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": 8.28774356842041, | |
| "smashed_token_generation_latency_sync": 55.83218421936035, | |
| "smashed_token_generation_latency_async": 55.261676385998726, | |
| "smashed_token_generation_throughput_sync": 0.0179108163863172, | |
| "smashed_token_generation_throughput_async": 0.018095723209970577, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 199.42901458740235, | |
| "smashed_inference_latency_async": 112.66343593597412, | |
| "smashed_inference_throughput_sync": 0.005014315505037694, | |
| "smashed_inference_throughput_async": 0.008875994165208075, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |