Model Card for llava-polyglot-ko-1.3b-hf

Model Description

llava-polyglot-ko-1.3b-hf is a model based on polyglot-ko-13b. We use llava for the vision question answering. You can see ‘demo.py’ and ‘llava_gpt_neox.py’. Currently, the model has been trained on small vision question answer dataset (approx, 10k) with 1.3b (small) model.

TODO

• Multi-turn chat based on the image
• Larger LLM
• More pretraining on for the vision-text adapter

References

• LLaVA
• polyglot korean