Training is now GA!

Since launching the beta of Baseten Training in May, we’ve introduced a ton of improvements, including;

• A more robust ML Cookbook, great starting points for:

  • Training a coding model with GRPO

  • Long-context training using multi-node with Qwen3 30B A3B

  • A variety of examples with Qwen3, gpt-oss, Gemma3, and Llama

• Resume from checkpoint: Launch jobs that pick up right where you left off

• A ton of other improvements, including:

  • Broader checkpoint recognition across FSDP, VeRL, and Megatron checkpointing formats

  • More availability for InfiniBand-backed multi-node training runs

  • Improved management and handling of the training cache

  • Per-GPU metric visibility and improved logs

  • Quality of life improvements around Training Cache

  • And much more!

After months of positive feedback from early users and thousands of training runs completed, Baseten Training is now immediately available for anyone on Baseten. Get started here.