Is it possible to try to support llm of other architectures as the backbone?

VectorSpaceLab / OmniGen

OmniGen: Unified Image Generation. https://arxiv.org/pdf/2409.11340

MIT License

2.57k stars 190 forks source link

Is it possible to try to support llm of other architectures as the backbone? #58

Open win10ogod opened 2 weeks ago

win10ogod commented 2 weeks ago

Is it possible to try to support llm of other architectures as the backbone? Perhaps the model definition can be updated, or can we only use phi3 architecture models?

staoxiao commented 2 weeks ago

You can replace phi3 with any other models. But note that you might need to modify the model code.