| { | |
| "base_current_gpu_type": "NVIDIA A100-PCIE-40GB", | |
| "base_current_gpu_total_memory": 40339.3125, | |
| "base_perplexity": 11.945297241210938, | |
| "base_token_generation_latency_sync": 56.39304389953613, | |
| "base_token_generation_latency_async": 49.49269890785217, | |
| "base_token_generation_throughput_sync": 0.01773268351645451, | |
| "base_token_generation_throughput_async": 0.020205000375143146, | |
| "base_token_generation_CO2_emissions": null, | |
| "base_token_generation_energy_consumption": null, | |
| "base_inference_latency_sync": 122.70581817626953, | |
| "base_inference_latency_async": 39.78419303894043, | |
| "base_inference_throughput_sync": 0.008149572814579, | |
| "base_inference_throughput_async": 0.02513561099558331, | |
| "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": 12.72653579711914, | |
| "smashed_token_generation_latency_sync": 167.0605026245117, | |
| "smashed_token_generation_latency_async": 167.2410013154149, | |
| "smashed_token_generation_throughput_sync": 0.005985855329596479, | |
| "smashed_token_generation_throughput_async": 0.005979394957783168, | |
| "smashed_token_generation_CO2_emissions": null, | |
| "smashed_token_generation_energy_consumption": null, | |
| "smashed_inference_latency_sync": 266.14876251220704, | |
| "smashed_inference_latency_async": 195.19712924957275, | |
| "smashed_inference_throughput_sync": 0.0037572971993590784, | |
| "smashed_inference_throughput_async": 0.005123026162548898, | |
| "smashed_inference_CO2_emissions": null, | |
| "smashed_inference_energy_consumption": null | |
| } |