| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 79.16311569213867, | |
| "base_token_generation_latency_async": 79.15753591805696, | |
| "base_token_generation_throughput_sync": 0.012632145554868622, | |
| "base_token_generation_throughput_async": 0.012633035988325728, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 77.93448944091797, | |
| "base_inference_latency_async": 77.57904529571533, | |
| "base_inference_throughput_sync": 0.01283128955066933, | |
| "base_inference_throughput_async": 0.012890078708602383, | |
| "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": 152.54314575195312, | |
| "smashed_token_generation_latency_async": 152.70157847553492, | |
| "smashed_token_generation_throughput_sync": 0.006555522341371384, | |
| "smashed_token_generation_throughput_async": 0.006548720779335068, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 153.68970184326173, | |
| "smashed_inference_latency_async": 153.36310863494873, | |
| "smashed_inference_throughput_sync": 0.00650661682602414, | |
| "smashed_inference_throughput_async": 0.00652047294098809, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |