train Kronos Large model

Since the Kronos-Large model (499.2M parameters) weights aren't publicly available, our team plans to train it from scratch. Preliminary estimates show we'll need 4–8 NVIDIA A100/H100 GPUs over a 2- to 3-month period for a practical turnaround.

If you have access to GPU capacity or compute credits to support this, our team can handle the end-to-end execution—from data pipeline preparation to distributed training and validation.

Let me know if you'd be interested in collaborating or providing sponsorship。