| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.633415222167969, | |
| "base_token_generation_latency_sync": 34.2150146484375, | |
| "base_token_generation_latency_async": 34.84920393675566, | |
| "base_token_generation_throughput_sync": 0.029226934732458666, | |
| "base_token_generation_throughput_async": 0.028695060059759192, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 122.62328338623047, | |
| "base_inference_latency_async": 45.04399299621582, | |
| "base_inference_throughput_sync": 0.008155058096513923, | |
| "base_inference_throughput_async": 0.022200518503854905, | |
| "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": 9.383063316345215, | |
| "smashed_token_generation_latency_sync": 167.0510223388672, | |
| "smashed_token_generation_latency_async": 166.85648560523987, | |
| "smashed_token_generation_throughput_sync": 0.005986195031907527, | |
| "smashed_token_generation_throughput_async": 0.005993174292103133, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 267.10435943603517, | |
| "smashed_inference_latency_async": 197.52447605133057, | |
| "smashed_inference_throughput_sync": 0.00374385503146187, | |
| "smashed_inference_throughput_async": 0.005062663726494992, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |