Skywork-13B

Training compute

2.5×10²³ FLOP

Parameters

13B

Published

Oct 30, 2023

Skywork-13B is an AI model developed by Kunlun Inc. (China), first published in October 2023. It works in the language domain, on tasks such as language modeling, language modeling/generation and translation.

Training it took an estimated 2.5×10²³ FLOP of compute (estimation method: operation counting). The model has 13,000,000,000 parameters. It was trained on roughly 3.2T datapoints. Training ran on 512 NVIDIA A800 PCIe 40 GB for about 940 hours.

Access: Open weights (restricted use). Its weights are openly available. The reference paper has 129 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Kunlun Inc.
Country of organization: China
Domain: Language
Task: Language modeling, Language modeling/generation, Translation
Training compute: 2.5×10²³ FLOP
Compute estimation method: Operation counting
Parameters: 13,000,000,000
Dataset size: 3.2T
Training hardware: NVIDIA A800 PCIe 40 GB
Chips used: 512
Training time: 940 h
Training power draw: 254.0 kW
Numerical format: BF16
Model accessibility: Open weights (restricted use)
Open weights: Yes
Citations: 129
Epoch confidence: Confident