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

Source: HuggingFace Date: 2025-09-10 URL: https://huggingface.co/blog/togethercomputer/together-ft

Summary

Integration announcement: Together AI enables fine-tuning of any compatible CausalLM model from the HF Hub (under 100B parameters) via their inference infrastructure, with automatic push-back to the Hub. A “base model template” approach handles architecture mapping; the workflow is a few lines of Python. Models stay in the HF ecosystem throughout — pull, fine-tune, push.

Implications

Thread: HF as open-source ML hub. Together AI’s bidirectional HF integration is the key feature: it positions HF Hub as the source-of-truth for both base and fine-tuned models, rather than an export destination. The “any CausalLM under 100B” claim is ambitious — architecture compatibility at that breadth will have real limits in practice. But the pattern (cloud fine-tuning provider as HF Hub citizen rather than a separate platform) continues the consolidation of the training + deployment stack around the Hub as the common substrate.