-
Notifications
You must be signed in to change notification settings - Fork 295
Open
Labels
bugSomething isn't workingSomething isn't workingqwenFor any PR / issue related to Qwen supportFor any PR / issue related to Qwen support
Description
⚙️ Your current environment
The output of python collect_env.py
Your output of `python collect_env.py` here
🐛 Describe the bug
- On transformers v4.57.1,
examples/awq/qwen3_moe_example.pyworks - After
uv pip install git+https://github.com/huggingface/transformers@main, it fails with last logs:
Preparing cache: 100%|██████████████████████████████████████████████| 256/256 [00:00<00:00, 353.13it/s]
(1/49): Calibrating: 0%| | 0/256 [00:02<?, ?it/s]
Traceback (most recent call last):
File "/home/brian-dellabetta/projects/llm-compressor/src/llmcompressor/pipelines/sequential/helpers.py", line 73, in forward
outputs = forward_fn(*args, **kwargs)
File "<string>", line 20, in forward
File "/home/brian-dellabetta/projects/.venv/lib/python3.10/site-packages/transformers/modeling_layers.py", line 94, in __call__
return super().__call__(*args, **kwargs)
File "/home/brian-dellabetta/projects/.venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/home/brian-dellabetta/projects/.venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1786, in _call_impl
return forward_call(*args, **kwargs)
File "/home/brian-dellabetta/projects/.venv/lib/python3.10/site-packages/transformers/models/qwen3_moe/modeling_qwen3_moe.py", line 336, in forward
hidden_states = residual + hidden_states
TypeError: unsupported operand type(s) for +: 'Tensor' and 'tuple'
This happens with oneshot(..., moe_calibrate_all_experts=False) as well
🛠️ Steps to reproduce
uv pip install git+https://github.com/huggingface/transformers@main, run examples/awq/qwen3_moe_example.py
Metadata
Metadata
Assignees
Labels
bugSomething isn't workingSomething isn't workingqwenFor any PR / issue related to Qwen supportFor any PR / issue related to Qwen support