A collection of end-to-end Alpamayo recipes for multiple versions (v1, v1.5, and beyond), designed to help developers quickly build, adapt, and deploy Alpamayo-based applications. This repo brings together battle-tested workflows across the Alpamayo ecosystem, including post-training recipes (supervised fine-tuning and reinforcement learning), quantization recipes, etc. Whether you are experimenting locally or building a full production stack, this repository is intended as the primary starting point for developers to learn, customize, and extend Alpamayo for their own use cases.
See CONTRIBUTING.md for the repository layout, recipe packaging conventions, and guidance on adding new recipes for released Alpamayo models.
Each recipe folder contains its own README with installation and training instructions.
| Recipe | Description |
|---|---|
recipes/alpamayo1_sft/ |
Alpamayo 1 supervised fine-tuning (HuggingFace Trainer + DeepSpeed) |
recipes/alpamayo1_5_sft/ |
Alpamayo 1.5 SFT (HuggingFace Trainer + DeepSpeed) |
recipes/alpamayo1_x_rl/ |
Alpamayo 1 and 1.5 RL post-training (Cosmos-RL / GRPO) |
| Script | Purpose |
|---|---|
scripts/curate_pai_samples.py |
Curate a subset of PAI samples |
scripts/convert_checkpoint.py |
Convert between Alpamayo 1 and 1.5 checkpoints |
scripts/convert_release_config_to_training.py |
Convert a release checkpoint to training format |
scripts/convert_cosmos_rl_checkpoint.py |
Convert a Cosmos-RL checkpoint to HuggingFace format |
scripts/download_pai.py |
Download the Physical AI AV dataset from HuggingFace |