FragLlama: Next-fragment prediction for molecular design

Training compute

3.3×10²³ FLOP

Parameters

7B

Published

May 16, 2024

FragLlama: Next-fragment prediction for molecular design is an AI model developed by Facebook AI Research (United States and France), first published in May 2024. It works in the multimodal, image generation, vision and language domain, on tasks such as language modeling/generation, vision-language generation, visual question answering and text-to-image.

Training it took an estimated 3.3×10²³ FLOP of compute (estimation method: hardware,operation counting). The model has 7,000,000,000 parameters. It was trained on roughly 4.4T datapoints. Training ran on NVIDIA A100 SXM4 80 GB for about 836 hours.

Access: Open weights (non-commercial). Its weights are openly available. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Facebook AI Research
Country of organization: United States, France
Domain: Multimodal, Image generation, Vision, Language
Task: Language modeling/generation, Vision-language generation, Visual question answering, Text-to-image
Training compute: 3.3×10²³ FLOP
Compute estimation method: Hardware, Operation counting
Parameters: 7,000,000,000
Dataset size: 4.4T
Training hardware: NVIDIA A100 SXM4 80 GB
Training time: 836 h
Model accessibility: Open weights (non-commercial)
Open weights: Yes
Epoch confidence: Confident