veri-examples

Runnable example scripts for the Veri RL post-training platform.

Contents

• quickstart.py — End-to-end GRPO training: connect a HuggingFace dataset, upload a reward function, submit a job, wait, download the checkpoint.
• math_reward_trl.py — TRL-format reward function for math reasoning (def reward(completions, answer, **kwargs) -> list[float]).
• math_reward.py — Miles-format async reward function (async def reward(args, sample, **kwargs) -> float).
• gsm8k_prompts.jsonl — 20-row sample dataset in Veri's chat format, useful for smoke tests without pulling the full GSM8K from HuggingFace.

Run

pip install veri-sdk
veri login --key vk_your_key
python quickstart.py

Docs

Full documentation: docs.veri.studio.