fast_pos_embed_interpolate function not handling pos embeds properly for batch size > 1
#15
by
jvv7
- opened
When I use batch size except 1, I am getting the following error:
Traceback (most recent call last):
.venv/lib/python3.11/site-packages/transformers/models/qwen3_vl/modeling_qwen3_vl.py", line 717, in forward
hidden_states = hidden_states + pos_embeds
~~~~~~~~~~~~~~^~~~~~~~~~~~
RuntimeError: The size of tensor a (batch_size) must match the size of tensor b (8192) at non-singleton dimension 0