[Question] lora fine-tuning inference

[pranavsingapore]

Question

I finetuned the llava-v1.5-13b model (from the repository liuhaotian/llava-v1.5-13b) using LoRA for 10 epochs. After merging the weights with the provided scripts, I uploaded the weights on huggingface and used the script https://github.com/LLaVA-VL/LLaVA-NeXT/blob/inference/docs/LLaVA-NeXT.md to do the inference, but the model's output during inference is "na".

Any guidance on what might be going wrong would be appreciated. Thank you!