KeyError: 'model.layers.17.mlp.experts.gate_up_proj'

#144
by pnaik24 - opened

While running the pipeline in the model card, I am getting the error: KeyError: 'model.layers.17.mlp.experts.gate_up_proj'.

same here. any solution?

Looks like it was OOM issue. I switched to a higher memory machine and it worked

I tried this, and it worked:

Install the correct stack (MXFP4 enabled)

In your virtual environment on Ubuntu (not Windows):

pip uninstall -y triton kernels transformers accelerate torch
pip install -U \
  "torch" \
  "transformers==4.57.1" \
  "accelerate" \
  "triton==3.4" \
  "kernels"

These are exactly the packages recommended in the official guide for GPT-OSS with MXFP4.

Sign up or log in to comment