Qwen/Qwen2.5-VL-3B-Instruct · Dimension out of range (expected to be in range of [-3, 2], but got 3)

I am trying to run the model locally on macos/m4 by strictly following the instructions in README.

I am getting this error when trying to run the first Python code example:

 File "/Users/john/work/venv/lib/python3.13/site-packages/transformers/models/qwen2_5_vl/modeling_qwen2_5_vl.py", line 272, in forward
    attn_output = F.scaled_dot_product_attention(q, k, v, attention_mask, dropout_p=0.0)
IndexError: Dimension out of range (expected to be in range of [-3, 2], but got 3)

Any ideas on how to fix it?

P.S. the only diverge from the README I have is that I needed to change inputs = inputs.to("cuda") to inputs = inputs.to("mps") to make it pass this point.