VASA-1

Training compute

4×10¹⁹ FLOP

Parameters

229M

Published

Oct 31, 2024

VASA-1 is an AI model developed by Microsoft Research Asia (China), first published in October 2024. It works in the video and audio domain, on tasks such as video generation.

Training it took an estimated 4×10¹⁹ FLOP of compute (estimation method: hardware). The model has 229,000,000 parameters. Training ran on 4 NVIDIA RTX A6000 for about 240 hours.

Access: Unreleased. Its weights are not openly released. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Microsoft Research Asia
Country of organization: China
Domain: Video, Audio
Task: Video generation
Training compute: 4×10¹⁹ FLOP
Compute estimation method: Hardware
Parameters: 229,000,000
Training hardware: NVIDIA RTX A6000
Chips used: 4
Training time: 240 h
Training power draw: 2.4 kW
Model accessibility: Unreleased
Open weights: No
Epoch confidence: Confident