Together AI expands fine-tuning service with more models, live metrics, and finer controls

Summary

Together AI is expanding its Fine-Tuning service to cover newer open-weight models and more of the workflow from dataset inspection through model selection. The update adds live training and evaluation metrics through the API, CLI, dashboard, and Python SDK, including curves that can be compared across in-flight jobs and underlying series that can be exported. Training controls now include Expert LoRA for mixture-of-experts layers, early stopping based on validation loss, and arbitrary effective batch sizes through gradient accumulation. The post reports that Expert LoRA recalled up to 89% of 200 invented facts versus 15% for attention-only adapters, while listed LoRA training prices fall by 30% to 70% for selected models, including the gpt-oss series.

Takeaways

1. 1

   Expert LoRA attaches adapters to mixture-of-experts layers instead of limiting them to attention layers. In the reported test, it recalled up to 89% of 200 invented facts, compared with 15% for attention-only adapters.

2. 2

   Early stopping monitors validation loss, keeps the checkpoint with the best result, and refunds unused training steps. Patience, sensitivity to small improvements, and warmup can be tuned.

3. 3

   Gradient accumulation allows an effective batch size of batch_size multiplied by gradient_accumulation_steps, enabling jobs whose desired batch size does not fit in GPU memory in one pass.

FAQ