| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 11.575923919677734, | |
| "base_token_generation_latency_sync": 35.51417636871338, | |
| "base_token_generation_latency_async": 34.99199263751507, | |
| "base_token_generation_throughput_sync": 0.028157769720402737, | |
| "base_token_generation_throughput_async": 0.028577966689667614, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 117.8853370666504, | |
| "base_inference_latency_async": 38.33281993865967, | |
| "base_inference_throughput_sync": 0.008482819194338112, | |
| "base_inference_throughput_async": 0.02608730590653659, | |
| "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": 11.824755668640137, | |
| "smashed_token_generation_latency_sync": 58.55119590759277, | |
| "smashed_token_generation_latency_async": 59.65384058654308, | |
| "smashed_token_generation_throughput_sync": 0.01707907045277486, | |
| "smashed_token_generation_throughput_async": 0.016763380029978882, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 193.11810760498048, | |
| "smashed_inference_latency_async": 108.6493730545044, | |
| "smashed_inference_throughput_sync": 0.0051781783303587545, | |
| "smashed_inference_throughput_async": 0.009203918733137521, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |