RDT-1B

Training compute

4.1×10²² FLOP

Parameters

1.2B

Published

Oct 10, 2024

RDT-1B is an AI model developed by Tsinghua University (China), first published in October 2024. It works in the robotics domain, on tasks such as robotic manipulation.

Training it took an estimated 4.1×10²² FLOP of compute. The model has 1,200,000,000 parameters. Training ran on 48 NVIDIA H100 PCIe for about 720 hours.

Access: Open weights (unrestricted). Its weights are openly available. The reference paper has 579 citations. Epoch AI rates the confidence of this record as likely.

Full record

Organization: Tsinghua University
Country of organization: China
Domain: Robotics
Task: Robotic manipulation
Training compute: 4.1×10²² FLOP
Parameters: 1,200,000,000
Training hardware: NVIDIA H100 PCIe
Chips used: 48
Training time: 720 h
Training power draw: 33.1 kW
Numerical format: BF16
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Citations: 579
Epoch confidence: Likely