deepmd-finetune-dpa3

What problem does it solve?

Fine-tune a pre-trained DPA3 model on a downstream dataset to adapt to new tasks.

Core Features & Use Cases

• Supports fine-tuning from a self-trained single-task DPA3 model, a multi-task pre-trained model, or built-in pretrained models downloaded via dp pretrained download.
• Handles single-task and multi-task workflows, including choosing appropriate model branches and data configurations.
• Provides guidance for preparing input.json, selecting learning rates, and freezing/testing the fine-tuned model.

Quick Start

Prepare downstream data and run the fine-tuning command with a chosen pretrained DPA3 model and an input.json.