fix: derive model_dir from MODEL_PATH for config loading

feat: HF Spaces Docker deployment setup

Fix tokenizer auto-detection in SFT and DPO training to support GPT-2 vocab size

Fix LoRA unwrapping in SFT final save so output model can be served directly

Fix API test compatibility with httpx 0.28+, fix model saving with tied weights, and update walkthrough documentation

upgrading the model

Support loading model.pt in from_pretrained

Support HuggingFace tokenizer in api_server.py

Fix imports: add sys.path for Kaggle/Colab compatibility

Update pipeline.py

Update modeling_openmind.py

Initial commit