| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 10.554253578186035, | |
| "base_token_generation_latency_sync": 38.07420387268066, | |
| "base_token_generation_latency_async": 37.691137939691544, | |
| "base_token_generation_throughput_sync": 0.026264501901181674, | |
| "base_token_generation_throughput_async": 0.02653143562818586, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 119.51892547607422, | |
| "base_inference_latency_async": 38.97833824157715, | |
| "base_inference_throughput_sync": 0.008366875756427245, | |
| "base_inference_throughput_async": 0.02565527534299363, | |
| "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": 357458.53125, | |
| "smashed_token_generation_latency_sync": 168.46778411865233, | |
| "smashed_token_generation_latency_async": 168.45296267420053, | |
| "smashed_token_generation_throughput_sync": 0.00593585298952883, | |
| "smashed_token_generation_throughput_async": 0.005936375259448941, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 262.0371948242188, | |
| "smashed_inference_latency_async": 209.70721244812012, | |
| "smashed_inference_throughput_sync": 0.0038162521189819085, | |
| "smashed_inference_throughput_async": 0.0047685532048517025, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |