v0.2.0

The first public release of loops-cookbook: post-train models on Baseten Loops. Loops runs the LoRA trainer and a paired sampler that hot-swaps to each new policy; the cookbook runs the rest.

What's in it

• Two recipes, one loops command. RL (recipe = "rl") for agents that act on tasks and get rewarded by a verifier, and SFT (recipe = "sft") for conversations the model should imitate.
• Serverless or dedicated compute. Every config picks [compute] scope. Serverless starts in minutes on a shared pool; dedicated gets its own trainer and sampler deployment (contact Baseten for access).
• Runtimes. The Harbor runtime on Docker, Blaxel or Daytona sandboxes, or your own custom runtime.
• Examples, patterns and tasks ship inside the package. Run any of them by path, e.g. uv run loops check examples/terminal_bench/train.toml.

Install

uv add baseten-loops-cookbook             # SFT, or RL with a custom runtime
uv add "baseten-loops-cookbook[harbor]"   # RL with the Harbor runtime

You need a Baseten workspace with Loops enabled and its API key. See the README for the quickstarts and the full list of extras.

Authors: @XavierFPMorris, @yorhodes

Full Changelog: https://github.com/basetenlabs/loops-cookbook/commits/v0.2.0