| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 31.111483764648437, | |
| "base_token_generation_latency_async": 31.104715168476105, | |
| "base_token_generation_throughput_sync": 0.03214247213552337, | |
| "base_token_generation_throughput_async": 0.03214946655462309, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 30.498201179504395, | |
| "base_inference_latency_async": 29.242229461669922, | |
| "base_inference_throughput_sync": 0.032788819055729314, | |
| "base_inference_throughput_async": 0.03419711897517179, | |
| "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_token_generation_latency_sync": 71.86500396728516, | |
| "smashed_token_generation_latency_async": 71.42011970281601, | |
| "smashed_token_generation_throughput_sync": 0.013914978707233166, | |
| "smashed_token_generation_throughput_async": 0.014001656734279755, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 70.03463668823242, | |
| "smashed_inference_latency_async": 69.58329677581787, | |
| "smashed_inference_throughput_sync": 0.014278649069768433, | |
| "smashed_inference_throughput_async": 0.014371265035368772, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |