| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.891793251037598, | |
| "base_token_generation_latency_sync": 39.88294296264648, | |
| "base_token_generation_latency_async": 39.48240038007498, | |
| "base_token_generation_throughput_sync": 0.02507337537594903, | |
| "base_token_generation_throughput_async": 0.02532774072431158, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 117.81007385253906, | |
| "base_inference_latency_async": 38.69774341583252, | |
| "base_inference_throughput_sync": 0.008488238461268461, | |
| "base_inference_throughput_async": 0.025841300079292664, | |
| "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": 285619.03125, | |
| "smashed_token_generation_latency_sync": 175.50264587402344, | |
| "smashed_token_generation_latency_async": 175.36316942423582, | |
| "smashed_token_generation_throughput_sync": 0.005697919794997303, | |
| "smashed_token_generation_throughput_async": 0.00570245167946763, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 266.68011779785155, | |
| "smashed_inference_latency_async": 230.2292823791504, | |
| "smashed_inference_throughput_sync": 0.003749810853008616, | |
| "smashed_inference_throughput_async": 0.004343496142915312, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |