DiffPepBuilder

Training compute

7.7×10²¹ FLOP

Parameters

104M

Published

Apr 30, 2024

DiffPepBuilder is an AI model developed by Peking University (China), first published in April 2024. It works in the biology domain, on tasks such as protein generation.

Training it took an estimated 7.7×10²¹ FLOP of compute (estimation method: hardware). The model has 104,000,000 parameters. It was trained on roughly 1.5M datapoints. Training ran on 8 NVIDIA A800 PCIe 40 GB for about 120 hours.

It is built on top of ESM2-650M. The reference paper has 27 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Peking University
Country of organization: China
Domain: Biology
Task: Protein generation
Training compute: 7.7×10²¹ FLOP
Compute estimation method: Hardware
Parameters: 104,000,000
Dataset size: 1.5M
Training hardware: NVIDIA A800 PCIe 40 GB
Chips used: 8
Training time: 120 h
Training power draw: 4.0 kW
Base model: ESM2-650M
Citations: 27
Epoch confidence: Confident