OpenFlamingo-9B-deprecated
| Property | Value |
|---|---|
| Base Architecture | CLIP ViT-L/14 + LLaMA-7B |
| Training Data | LAION 2B + Multimodal C4 |
| License | Non-commercial research only |
| Status | Deprecated (Superseded by v2) |
What is OpenFlamingo-9B-deprecated?
OpenFlamingo-9B is an open-source implementation of DeepMind's Flamingo visual-language model, combining CLIP's vision capabilities with LLaMA's language understanding. While now deprecated in favor of newer versions, it represents a significant milestone in multimodal AI development.
Implementation Details
The model employs a frozen pretrained vision encoder and language model architecture, enhanced with trainable Perceiver modules and cross-attention layers. It has been trained on 5 million interleaved image-text examples from Multimodal C4 and LAION 2B dataset.
- Utilizes CLIP ViT-L/14 for vision processing
- Integrates LLaMA-7B for language understanding
- Implements Perceiver modules for cross-modal interaction
- Demonstrates strong few-shot learning capabilities
Core Capabilities
- COCO captioning with CIDEr scores reaching 84.52 in 32-shot scenarios
- VQAv2 accuracy of up to 50.34 in 32-shot settings
- Progressive performance improvement with increased shot count
- Effective zero-shot to few-shot learning adaptation
Frequently Asked Questions
Q: What makes this model unique?
The model stands out for its implementation of DeepMind's Flamingo architecture in an open-source format, combining powerful vision and language models while demonstrating strong few-shot learning capabilities.
Q: What are the recommended use cases?
The model is strictly intended for academic research purposes only, specifically in areas like image captioning and visual question answering. Commercial use is prohibited due to LLaMA's licensing restrictions.
