| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 53.73468589782715, | |
| "base_token_generation_latency_async": 53.75340115278959, | |
| "base_token_generation_throughput_sync": 0.018609953390281875, | |
| "base_token_generation_throughput_async": 0.01860347398590803, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 52.639743423461915, | |
| "base_inference_latency_async": 51.52573585510254, | |
| "base_inference_throughput_sync": 0.018997053081271153, | |
| "base_inference_throughput_async": 0.019407777170075507, | |
| "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_token_generation_latency_sync": 162.5103317260742, | |
| "smashed_token_generation_latency_async": 162.95767445117235, | |
| "smashed_token_generation_throughput_sync": 0.006153454918088469, | |
| "smashed_token_generation_throughput_async": 0.006136562781519283, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 169.10120849609376, | |
| "smashed_inference_latency_async": 126.91173553466797, | |
| "smashed_inference_throughput_sync": 0.00591361829340859, | |
| "smashed_inference_throughput_async": 0.007879491961772393, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |