| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_token_generation_latency_sync": 101.87378540039063, | |
| "base_token_generation_latency_async": 101.67951695621014, | |
| "base_token_generation_throughput_sync": 0.009816067951826256, | |
| "base_token_generation_throughput_async": 0.009834822488689296, | |
| "base_token_generation_CO2_emissions": 2.6819935676011093e-05, | |
| "base_token_generation_energy_consumption": 0.008565413871084828, | |
| "base_inference_latency_sync": 90.84702758789062, | |
| "base_inference_latency_async": 88.71569633483887, | |
| "base_inference_throughput_sync": 0.011007514792187808, | |
| "base_inference_throughput_async": 0.01127196247466411, | |
| "smashed_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "smashed_current_gpu_total_memory": 40339.3125, | |
| "smashed_token_generation_latency_sync": 63.452283477783205, | |
| "smashed_token_generation_latency_async": 64.33862000703812, | |
| "smashed_token_generation_throughput_sync": 0.015759874116274063, | |
| "smashed_token_generation_throughput_async": 0.01554276420430852, | |
| "smashed_token_generation_CO2_emissions": 1.9680748845555618e-05, | |
| "smashed_token_generation_energy_consumption": 0.0053972414185424426, | |
| "smashed_inference_latency_sync": 76.47395782470703, | |
| "smashed_inference_latency_async": 73.17061424255371, | |
| "smashed_inference_throughput_sync": 0.013076346882584418, | |
| "smashed_inference_throughput_async": 0.013666688606509356 | |
| } |