Evaluation setup
What the accuracy figures on the model table were produced with: the SDK that quantized and compiled each model, and the machine the evaluation ran on.
Run
- Evaluated
- 2026-10-06
- Enlight SDK
- 3.1.3
- Models published
- 87
- Tasks
- 5
Machine
- CPU
- AMD EPYC 7742 64-Core Processor (128 threads)
- Memory
- 503.7 GiB
- GPU
- 4 × NVIDIA A100-SXM4-40GB
- OS
- Ubuntu 22.04.5 LTS
The machine is the host the evaluation ran on. Accuracy is measured on the FP32 model and on the quantized graph, and inference time comes from the Enlight simulator, so neither figure is measured on this hardware.