Supported Models

Train open source models on ART.

Serverless Models

We currently only support the following model for serverless training. We are actively adding support for both larger and smaller models. If there’s a particular model you’d like to see serverless support for, please send a request to [e-mail].

• OpenPipe Qwen 3 14B Instruct
  • Good balance of performance and size. Has support for tool calling and generally trains well. This is our recommended model for users new to RL.

Recommended Local Models

If you’re developing locally or in your own hardware, here are a couple other models you could try in addition to the recommended serverless list.

• Qwen2.5 7B Instruct
  • Less capable than 14B, but smaller and faster
• Qwen2.5 32B Instruct
  • More capable than 14B, but larger and slower

More Models

ART has wide support for models supported by vLLM. However, not all models support all features. For instance, if a model’s chat template does not include tool call support, you won’t be able to use tools with it natively. And if a model’s architecture doesn’t have support for LoRA layers, it won’t be compatible with our LoRA-based backend, but still may work with our full-fine-tuning backend.

Here are additional models that we’ve tested and found to work well with ART:

• Llama 3.1 8B Instruct
• Llama 3.2 1B Instruct
• Llama 3.2 3B Instruct
• Llama 3.3 70B Instruct
• Qwen2.5 72B Instruct
• Additionally, the Qwen 3 family of models is well supported for single-turn workflows. For multi-turn workflows the Qwen 3 chat template removes the <think> tokens from previous turns, which makes training more complicated. It is still possible to use for multi-turn workflows by splitting each turn into a separate message history with our additional_histories trajectory parameter (see Additional Histories).

If you’re curious about a model that is not listed above, ask in the Discord #support channel.