cannot import name 'draw_graph' from 'plot_utils'

[Iust1n2]

I got this error after i run the app with python app.py --model_name_or_path Intel/llava-gemma-2b --load_8bit:

Traceback (most recent call last):
  File "/Users//Desktop/lvlm-interpret/app.py", line 30, in <module>
    demo = build_demo(args, embed_mode=False)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users//Desktop/lvlm-interpret/utils_gradio.py", line 224, in build_demo
    processor, model = get_processor_model(args)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users//Desktop/lvlm-interpret/utils_model.py", line 44, in get_processor_model
    model = LlavaForConditionalGeneration.from_pretrained(
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/envs/attention-viz/lib/python3.12/site-packages/transformers/modeling_utils.py", line 3049, in from_pretrained
    hf_quantizer.validate_environment(
  File "/opt/anaconda3/envs/attention-viz/lib/python3.12/site-packages/transformers/quantizers/quantizer_bnb_8bit.py", line 62, in validate_environment
    raise ImportError(
ImportError: Using `bitsandbytes` 8-bit quantization requires Accelerate: `pip install accelerate` and the latest version of bitsandbytes: `pip install -i https://pypi.org/simple/ bitsandbytes`

I'm currently running accelerate version 0.30.1 and bitsandbytes version 0.42.0 on Mps with Python 3.10

[shaoyent-IL]

From the error message it looks like an import error with bitsandbytes I am also using bitsandbytes==0.42.0, accelerate==0.30.1, and transformers==4.39.3 You can check if your environment is set up correctly. Are you running on GPU?

[Iust1n2]

Needed to pull from causality labs submodule after initializing. Everything works properly on GPU. Btw, do you plan on releasing support to other VLMs from HuggingFace as in VL-Interpret? I would very much like to use this tool for interpretability purposes.