- Step1: Install ModelScope and Trinity-RFT via pip.
- Step2: Prepare and preprocess your text, image, or video corpus.
- Step3: Configure retrieval settings (vector store, encoder) in the YAML config.
- Step4: Run the retrieval index builder to generate embeddings and index files.
- Step5: Launch the fine-tuning script with retrieval integration enabled.
- Step6: Evaluate performance using provided metrics and visualizations.
- Step7: Deploy the retrieval-augmented model on ModelScope or export for serving.