| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 13.74486255645752, | |
| "base_token_generation_latency_sync": 38.00431823730469, | |
| "base_token_generation_latency_async": 38.87844365090132, | |
| "base_token_generation_throughput_sync": 0.026312799344428425, | |
| "base_token_generation_throughput_async": 0.02572119421701226, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 122.91768264770508, | |
| "base_inference_latency_async": 44.02568340301514, | |
| "base_inference_throughput_sync": 0.008135525975266752, | |
| "base_inference_throughput_async": 0.02271401424586436, | |
| "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": 14.683727264404297, | |
| "smashed_token_generation_latency_sync": 167.07373809814453, | |
| "smashed_token_generation_latency_async": 166.4058430120349, | |
| "smashed_token_generation_throughput_sync": 0.0059853811340030445, | |
| "smashed_token_generation_throughput_async": 0.006009404368857874, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 264.801692199707, | |
| "smashed_inference_latency_async": 193.66347789764404, | |
| "smashed_inference_throughput_sync": 0.003776410912230214, | |
| "smashed_inference_throughput_async": 0.005163596207481747, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |