ART
ART copied to clipboard
OpenPipe
Agent Reinforcement Trainer: train multi-step agents for real-world tasks using GRPO. Give your agents on-the-job training. Reinforcement learning for Qwen2.5, Qwen3, Llama, and more!
- Add exclude parameter to Backend._experimental_pull_from_s3 to skip specified dirs - Supported exclude values: checkpoints, logs, trajectories - Update API to accept exclude list and forward it to backend method...
```bash Art version: 0.4.4 Python: 3.12.11 ``` the function `art.rewards.ruler.ruler()` and the module itself `art.rewards.ruler` have the same name. Since in the `art.rewards.__init__` you import the function `art.rewards.ruler.ruler()` as `"ruler"`,...
This PR adds a GitHub Action workflow that prevents contributors from manually bumping the version in regular pull requests. ## What it does: - ✅ Runs on PRs that modify...
i get an issue during training where the process is getting stuck at the gather stage. specifically, at this progress point: gather: 75%|█████████████████████████████████████████████████████████████████████████████████████▌ | 9/12 [00:24
## Proposal * Create an `@art.rollout` decorator which wraps a rollout function and constructs a trajectory (potentially with multiple histories) automatically, similar to how `@weave.op` automatically wraps an LLM-enabled function...
As noted in the main READ.me, Gemma 3 models are not yet supported by ART, due to Gemma not accepting the enable_prefix_caching parameter. To solve this issue, I've introduced the...
### Changes * Add wikihop/wikihop.ipynb notebook in examples * Add wikihop/judge_group.py with generalized group judgement function