IDEFICS-80B

Training compute

1.2×10²³ FLOP

Parameters

80B

Published

Aug 22, 2023

IDEFICS-80B is an AI model developed by Hugging Face (United States), first published in August 2023. It works in the multimodal, language and vision domain, on tasks such as language modeling, image captioning and visual question answering.

Training it took an estimated 1.2×10²³ FLOP of compute (estimation method: hardware,operation counting). The model has 80,000,000,000 parameters. It was trained on roughly 149.6B datapoints. Training ran on 512 NVIDIA A100 for about 672 hours.

Access: Open weights (non-commercial). Its weights are openly available. It is built on top of LLaMA-65B,CLIP ViT-H/14 - LAION-2B. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Hugging Face
Country of organization: United States
Domain: Multimodal, Language, Vision
Task: Language modeling, Image captioning, Visual question answering
Training compute: 1.2×10²³ FLOP
Compute estimation method: Hardware, Operation counting
Parameters: 80,000,000,000
Dataset size: 149.6B
Training hardware: NVIDIA A100
Chips used: 512
Training time: 672 h
Training power draw: 407.0 kW
Numerical format: BF16
Model accessibility: Open weights (non-commercial)
Open weights: Yes
Base model: LLaMA-65B, CLIP ViT-H/14 - LAION-2B
Epoch confidence: Confident