Fine-tune Any LLM from the Hugging Face Hub with Together AI

Together AI and Hugging Face have announced a new capability that allows any compatible large language model (LLM) from the Hugging Face Hub to be fine‑tuned using Together AI’s infrastructure. The integration requires only a few API calls: users specify a base model that serves as a training template and a custom model that matches the base’s architecture, size, and sequence length. The system automatically allocates GPU resources, configures memory usage, and sets up the training pipeline. After training, the fine‑tuned model can be deployed for inference, downloaded, or uploaded back to the Hub, with private repositories accessible via an HF token. The feature supports all CausalLM models under 100B parameters and works bidirectionally, pulling public models and pushing fine‑tuned ones back to the Hub. Early adopters such as Slingshot AI and Parsed report accelerated development cycles, cost savings from fewer training epochs, and the ability to experiment with a wide range of model variants. The integration aims to reduce the friction of setting up fine‑tuning infrastructure and to enable teams to quickly iterate on community‑built models, thereby speeding time to value and leveraging collective open‑source intelligence. The platform is actively improving based on user feedback, and Together AI invites users to share experiences via Discord or direct outreach.