| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 80.28141708374024, | |
| "base_token_generation_latency_async": 80.70033676922321, | |
| "base_token_generation_throughput_sync": 0.012456182717314473, | |
| "base_token_generation_throughput_async": 0.012391522018794985, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 80.46366806030274, | |
| "base_inference_latency_async": 83.66153240203857, | |
| "base_inference_throughput_sync": 0.012427969344506634, | |
| "base_inference_throughput_async": 0.011952924734804799, | |
| "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": 145.91746520996094, | |
| "smashed_token_generation_latency_async": 145.90322263538837, | |
| "smashed_token_generation_throughput_sync": 0.0068531892228329075, | |
| "smashed_token_generation_throughput_async": 0.006853858207772397, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 143.4565643310547, | |
| "smashed_inference_latency_async": 147.71199226379395, | |
| "smashed_inference_throughput_sync": 0.006970751074815233, | |
| "smashed_inference_throughput_async": 0.006769931030475394, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |