some problem with the combine

[crazyCoder18]

The generated file is .satetensors, Instead of .bin file

[chengxin0913]

Same question. When I use the weight like this one for demo_app.py, the NExT-GPT can't generate any multi-modality object and might be some weights missing. I guess this is related to :

[Michael4933]

same...

[Michael4933]

Now I have figured it out. It is very likely that the llama weights you select is not appropriate. My former attempt failed at choose a incorrect .pth file at size 13G, and it turns out that three .safetensors files were generated, instead of 2 .bin files. Regarding the selection of llama, you are suggested to choose the ones on huggingface consisiting of two .bin file like this https://huggingface.co/yahma/llama-7b-hf/tree/main. This way, the combined output is likely to be .bin
Good Luck!