stavc / compromptmized Goto Github PK

Hello,

I am assuming that I can run the RAG-based worm without GPU?

I noticed that in your requirements.txt you have torch and torchvision(and potentially other packages) which requires GPU.

Hello,

When inspecting the model zoo of LLaVa (as suggested in the README of this repo), it is not clear what version of LLaVa-7B was used for this paper

You write in your REPO:

You can download the model checkpoints from the LLaVA repository and save them to the "models" folder. Save the weights in the "ComPromptMized/FlowSteering/llava/llava_weights" directory. In our experiments, we utilize the LLaVA-7B weights.

Which weights and checkpoints would you recommend?(To get your code running)

v/r

-Tulio

Hello,

I am having an issue when installing the requirements.txt file in the git repo:

ERROR: Could not find a version that satisfies the requirement pywin32==306 (from versions: none)

ERROR: No matching distribution found for pywin32==306

I am assuming this can only be run on a windows machine?

(I have read your paper : “Here Comes The AI Worm: Unleashing Zero-click

Worms that Target GenAI-Powered Applications”

but I may need to read it again)

If possible could you post a bigger dataset? I am able to run your code and I am currently using my own dataset for the RAG algorithm but I would like to see yours.