| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.891793251037598, | |
| "base_token_generation_latency_sync": 36.54602737426758, | |
| "base_token_generation_latency_async": 36.249685660004616, | |
| "base_token_generation_throughput_sync": 0.02736275518427784, | |
| "base_token_generation_throughput_async": 0.027586446110988778, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.13349075317383, | |
| "base_inference_latency_async": 39.10982608795166, | |
| "base_inference_throughput_sync": 0.00839394525987529, | |
| "base_inference_throughput_async": 0.025569021906442696, | |
| "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": 9.47966194152832, | |
| "smashed_token_generation_latency_sync": 167.53070373535155, | |
| "smashed_token_generation_latency_async": 167.87611786276102, | |
| "smashed_token_generation_throughput_sync": 0.005969055090819061, | |
| "smashed_token_generation_throughput_async": 0.005956773439432889, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 265.20934295654297, | |
| "smashed_inference_latency_async": 196.3787078857422, | |
| "smashed_inference_throughput_sync": 0.003770606226960335, | |
| "smashed_inference_throughput_async": 0.005092201750211249, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |