redundant llava_lama copies

[4lt3r3go]

There is some confusion regarding Llava_Lama .. i'm sorry its always me annoy i know 😁

When using your nodes, it is automatically downloaded and placed in the LLM folder
ComfyUI\models\LLM\llava-llama-3-8b-text-encoder-tokenizer

However, comfy native nodes now asking to download one of this

which are placed in a different folder: ComfyUI/models/text_encoders

As a result, many users end up with a sort of duplicate (or am I wrong?).

Is there a way to avoid this?
The files are quite large, and having duplicates seems a bit redundant.

[kijai]

It's not only about the files, comfy has his own code to use them so they wouldn't be compatible anyway. Sure I could work on utilizing that, but at this point it feels like wasted effort as eventually people will move onto the native only anyway.

[4lt3r3go]

i think your way has more sense tho. i'll stick on your nodes for now
admit that the ability to choose different samplers is inviting... uhm ok i'll give it a try

[Ratinod]

people will move onto the native only anyway

This transition is very painful without sageattn, BlockSwap and Torch Compile...