AI models
InstructBLIP
Training compute
1.9×10²⁰ FLOP
Parameters
13B
Published
May 11, 2023
InstructBLIP is an AI model developed by Salesforce Research, Hong Kong University of Science and Technology (HKUST) and Nanyang Technological University (United States, Hong Kong and Singapore), first published in May 2023. It works in the multimodal, language and vision domain, on tasks such as visual question answering and chat.
Training it took an estimated 1.9×10²⁰ FLOP of compute (estimation method: hardware). The model has 13,000,000,000 parameters. Training ran on 16 NVIDIA A100 SXM4 40 GB for about 36 hours.
Access: Open weights (non-commercial). Its weights are openly available. It is built on top of Vicuna-13B v0. The reference paper has 3,418 citations. Epoch AI rates the confidence of this record as confident.
Full record
- Organization
- Salesforce Research, Hong Kong University of Science and Technology (HKUST), Nanyang Technological University
- Country of organization
- United States, Hong Kong, Singapore
- Domain
- Multimodal, Language, Vision
- Task
- Visual question answering, Chat
- Training compute
- 1.9×10²⁰ FLOP
- Compute estimation method
- Hardware
- Parameters
- 13,000,000,000
- Training hardware
- NVIDIA A100 SXM4 40 GB
- Chips used
- 16
- Training time
- 36 h
- Chip-hours
- 576
- Training power draw
- 12.7 kW
- Numerical format
- FP16
- Model accessibility
- Open weights (non-commercial)
- Open weights
- Yes
- Base model
- Vicuna-13B v0
- Citations
- 3,418
- Epoch confidence
- Confident
SourceEpoch AI, 'AI Models'. Published online at epoch.ai. Retrieved 2026-07-29 from https://epoch.ai/data/ai-models. Licensed under CC BY 4.0.
← All ai models