Upload FastGen/README.md with huggingface_hub

FastGen/README.md

@@ -50,6 +50,56 @@ cd FastGen
 pip install -e .
 ```
 
+### Offline Installation
+
+For air-gapped or isolated environments without internet access, use the pre-packaged offline bundle from HuggingFace:
+
+```bash
+# On a machine WITH internet: download the offline package (~68 GB)
+pip install huggingface_hub
+huggingface-cli download taohu/fastgen-offline --local-dir ./offline_package
+
+# Transfer to offline machine (USB, rsync, etc.)
+# rsync -avz ./offline_package user@offline-machine:/path/to/
+```
+
+On the offline machine:
+
+```bash
+cd /path/to/offline_package
+
+# Run the setup script (creates conda env, installs packages, configures paths)
+bash FastGen/scripts/setup_offline_env.sh
+
+# Activate the environment
+source activate_fastgen.sh
+
+# Verify installation
+python -c "import torch; print(f'PyTorch: {torch.__version__}')"
+python -c "from diffusers import WanPipeline; print('Diffusers: OK')"
+python -c "from fastgen.networks.Wan.network import Wan; print('FastGen: OK')"
+```
+
+Run data-free training (no video dataset required):
+
+```bash
+cd FastGen
+python train.py --config=fastgen/configs/experiments/WanT2V/config_sf_datafree.py
+```
+
+The offline package includes:
+- Python wheels for all dependencies (~3.3 GB)
+- Wan-T2V-1.3B model from HuggingFace (~27 GB)
+- Self-Forcing checkpoints (~38 GB)
+- FastGen source code
+
+To prepare your own offline package (e.g., for different Python version or platform):
+
+```bash
+python scripts/prepare_offline_package.py --output-dir ./my_offline_package \
+    --python-version 311 --platform manylinux2014_x86_64 --create-archive
+```
+
 ### Credentials (Optional)
 
 For W&B logging, [get your API key](https://wandb.ai/settings) and save it to `credentials/wandb_api.txt` or set the `WANDB_API_KEY` environment variable.