InternLM-XComposer

Training compute

5.3×10²¹ FLOP

Parameters

7B

Published

Sep 26, 2023

InternLM-XComposer is an AI model developed by Shanghai AI Lab (China), first published in September 2023. It works in the multimodal, language and vision domain, on tasks such as chat, visual question answering and translation.

Training it took an estimated 5.3×10²¹ FLOP of compute (estimation method: hardware,operation counting). The model has 7,000,000,000 parameters. It was trained on roughly 125.6B datapoints. Training ran on 128 NVIDIA A100 for about 80 hours.

Access: Open weights (restricted use). Its weights are openly available. It is built on top of InternLM. The reference paper has 341 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Shanghai AI Lab
Country of organization: China
Domain: Multimodal, Language, Vision
Task: Chat, Visual question answering, Translation
Training compute: 5.3×10²¹ FLOP
Compute estimation method: Hardware, Operation counting
Parameters: 7,000,000,000
Dataset size: 125.6B
Training hardware: NVIDIA A100
Chips used: 128
Training time: 80 h
Chip-hours: 10.2K
Training power draw: 101.7 kW
Model accessibility: Open weights (restricted use)
Open weights: Yes
Base model: InternLM
Citations: 341
Epoch confidence: Confident