FabianMenekshi/ML_and_AI_project

This work tackles OpenAI’s Parameter Golf Challenge under strict limits: 5000 training iterations on one NVIDIA A100 GPU and a final model size under 16 MB. By adding architectural mechanisms and using mixed-bitwidth quantization with quantization-aware training, the final model reaches 1.2812 BpB at 15.36 MB.

语言：Jupyter Notebook · GitHub 星标：4 · Fork：0

在 GitHub 上查看原仓库