alpamayo-recipes
Fine-tune, quantize, and RL-train autonomous driving models
All Skills in This Repository (3)
Pure Emerald Level Indicators
NVlabsalpamayo1-5-sft
Automate Alpamayo-1.5 VLM fine-tuning on PAI or LingoQA datasets.
NVlabsalpamayo1-sft
Automate end-to-end fine-tuning of the Alpamayo-1 VLM on the PAI dataset.
NVlabsalpamayo1-x-rl
Automate RL post-training for Alpamayo VLM action models with Cosmos-RL and GRPO.
Frequently Asked Questions
FAQPage SchemaHow to install alpamayo-recipes?▼
Run `npx skills add NVlabs/alpamayo-recipes --all -g -y` in your terminal to install all skills in this suite globally.
What does alpamayo-recipes do?▼
It provides ready-to-run recipes for post-training NVIDIA Alpamayo autonomous driving models, covering supervised fine-tuning, reinforcement learning with GRPO, and FP8 quantization on the Physical AI AV dataset.
What hardware do I need to run these recipes?▼
The SFT recipes were validated on 8x H100 80GB GPUs, and the RL recipe needs at least 5 GPUs with 80GB each for a local test run.
Can I fine-tune Alpamayo on my own driving data?▼
Yes. You can point the recipes at your own fleet data, adjust camera counts and sensor configs, and swap reward functions for the RL pipeline.
Do I need deep learning expertise to use these skills?▼
No. The agent asks a few setup questions up front, then runs the full training, evaluation, and checkpoint export pipeline automatically.
Related Repositories in Software Engineering
View All in Software Engineering→
openclawopenclaw
Run a personal AI assistant across your devices and chat apps
obrasuperpowers
Gives coding agents a disciplined workflow from idea to merged code
facebookreact
AI agent skills for building, testing, and porting React core