Skip to content

Assertion Error When Running llama3.2-vision Model in ollama

Description:

The ollama package fails with an assertion error when attempting to run the llama3.2-vision model. The error message is as follows: Error: llama runner process has terminated: GGML_ASSERT(ggml_nelements(a) == ne0*ne1*ne2) failed

This issue appears to be related to upstream bug https://github.com/ollama/ollama/issues/7590 in the ollama GitHub repository.

Additional info:

Package version(s): 
    ollama 0.4.7-1, ollama-cuda 0.4.7-1
    cuda 12.6.3-1
    nvidia-open-dkms 565.57.01-2

Steps to reproduce:

  1. Install and set up the ollama package.
  2. Run the following command to load the llama3.2-vision model:

ollama run llama3.2-vision

To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information