LLaVA-NeXT
LLaVA-NeXT copied to clipboard
[Question] Is it possible to convert `LlavaMistralForCausalLM` to `LlavaNextForConditionalGeneration`?
This script(scripts/train/finetune_siglip_a4.sh) was used to train Mistal-v0.3 as a vision model.
The training seemed to work and the model was output, but the model's architecture is LlavaMistralForCausalLM
, so it cannot be inferred by transformers or vllm.
Is there any way to convert this model to LlavaNextForConditionalGeneration
or is there an alternative?