| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 8.633415222167969, | |
| "base_token_generation_latency_sync": 34.17437381744385, | |
| "base_token_generation_latency_async": 33.45911428332329, | |
| "base_token_generation_throughput_sync": 0.029261691972526017, | |
| "base_token_generation_throughput_async": 0.029887222702079136, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 120.59002914428712, | |
| "base_inference_latency_async": 43.94807815551758, | |
| "base_inference_throughput_sync": 0.008292559568117282, | |
| "base_inference_throughput_async": 0.022754123546912194, | |
| "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": 163207.5625, | |
| "smashed_token_generation_latency_sync": 174.02754669189454, | |
| "smashed_token_generation_latency_async": 173.97046349942684, | |
| "smashed_token_generation_throughput_sync": 0.005746216728380598, | |
| "smashed_token_generation_throughput_async": 0.005748102177145114, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 267.41893005371094, | |
| "smashed_inference_latency_async": 229.5799970626831, | |
| "smashed_inference_throughput_sync": 0.0037394510545650248, | |
| "smashed_inference_throughput_async": 0.004355780175948718, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |