Principle:Intel Ipex llm DPO Model Export
| Knowledge Sources | |
|---|---|
| Domains | NLP, Model_Deployment |
| Last Updated | 2026-02-09 00:00 GMT |
Overview
Process for saving trained DPO model artifacts and optionally merging LoRA adapters for deployment.
Description
After DPO training, the model artifacts (adapter weights + tokenizer) are saved using save_pretrained. Optionally, the LoRA adapter can be merged into the base model using the same merge_adapter utility as QLoRA/LoRA workflows. The DPO export workflow uses the same export_merged_model.py script as other finetuning workflows.
Usage
Use after DPO training completes. Save adapter separately for flexible deployment, or merge into base model for standalone deployment.
Practical Guide
- Save adapter + tokenizer with save_pretrained
- Optionally merge adapter into base model with merge_adapter/export_merged_model.py
- Merged model can be loaded with standard HuggingFace methods
Related Pages
Implemented By
Page Connections
Double-click a node to navigate. Hold to expand connections.
Principle
Implementation
Heuristic
Environment