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!
Added encoding="utf-8" to file read/write operations in bump_version.py to handle pyproject.toml with UTF-8 encoding, resolving 'gbk' codec can't decode byte 0xbf error at position 1779.
## Summary - Fixed critical resource cleanup bug in `LocalBackend.close()` that was causing service close methods to not execute properly when they were async - Made `LocalBackend.close()` async to properly...
This pull request refactors the TrajectoryGroup class to simplify its initialization and remove the need for monkey-patching in the ART CLI. Previously, the TrajectoryGroup class had a complex __new__ method...
README_zh_CN.md is the translated version of README.md in Chinese(Simplified). This close issue #271.
This pull request introduces a new "trash" feature to the ART framework, providing a safety net for model deletion. Previously, deleting a model was a destructive and irreversible operation. Now,...
## Summary - Added `adjust_lr` function to `iterate_dataset.py` to support learning rate schedules with warmup and cooldown phases - Updated `DatasetBatch` to include `total_steps` field needed for LR calculations -...
Migrate the Unsloth model service to also support vLLM V1 which has some performance improvements and is the future of vLLM development.