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!
Hey! It'd be absolutely huge for us if you could add multimodal support to the library. We do visual navigation and so it's a pre-requisite for us using the library.
I am using the RULER notebook to train a model, but I get this error: ``` --------------------------------------------------------------------------- RuntimeError Traceback (most recent call last) /usr/local/lib/python3.12/dist-packages/unsloth_zoo/vllm_utils.py in load_vllm(model_name, config, gpu_memory_utilization, max_seq_length, dtype,...
ref: https://github.com/meta-pytorch/torchtune/issues/2883 Candidates - torchforge - verl - TRL - prime-rl
How do I get ART to instantiate the model with FastModel instead of FastLanguageModel in Unsloth (Unsloth docs says: _If you're fine-tuning the MOE models, please use FastModel and not...
This pull request improves the scenario/test case generation process by using a local mcp (stdio) format server to generate the response needed for scenarios. Included in the examples directory are...
Enhance assertion for content type in message dictionary. Came across this when testing this out with a tool-call and the assertion / why it was failing was not obvious
Does ART currently support making connections to Azure OpenAI models? I see support for OpenAI models [here](https://github.com/OpenPipe/ART/blob/main/src/art/model.py#L73). It seems like it would only require a couple of updates in a...
Would you guys mind adding `endpoint` argument to the s3_sync in `art.utils.s3` ```python 65 @limit_concurrency(1) 66 async def s3_sync( 67 source: str, 68 destination: str, 69 *, 70 profile: Optional[str]...