| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 29282.24609375, | |
| "base_token_generation_latency_sync": 31.626212120056152, | |
| "base_token_generation_latency_async": 31.4815953373909, | |
| "base_token_generation_throughput_sync": 0.03161934145650777, | |
| "base_token_generation_throughput_async": 0.031764591002549775, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 129.1005958557129, | |
| "base_inference_latency_async": 31.828856468200684, | |
| "base_inference_throughput_sync": 0.007745897634102581, | |
| "base_inference_throughput_async": 0.03141803102474234, | |
| "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": 28594.048828125, | |
| "smashed_token_generation_latency_sync": 39.41377983093262, | |
| "smashed_token_generation_latency_async": 39.471565932035446, | |
| "smashed_token_generation_throughput_sync": 0.02537183706534999, | |
| "smashed_token_generation_throughput_async": 0.025334692870352827, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 183.90487060546874, | |
| "smashed_inference_latency_async": 65.73736667633057, | |
| "smashed_inference_throughput_sync": 0.00543759388594607, | |
| "smashed_inference_throughput_async": 0.015212048345709907, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |