[UT] tiny_qwen_moe_model_path loads the full Qwen1.5-MoE-A2.7B checkpoint
Open
- Dominant language
- Python
- Stars
- 1.6k
- Forks
- 175
- Avg merge
- 1d 18h
- Merged PRs (30d)
- 99
Description
## Problem
The following unit tests depend on `tiny_qwen_moe_model_path`:
- `test/unit/common/schemes/test_scheme.py::TestAutoRound::test_w4a16_mixed`
- `test/unit/test_cuda/algorithms/test_awq.py::TestAWQMoE::test_awq_moe_dynamic_smoothing`
- `test/unit/test_cuda/models/test_special_model.py::TestSpecialModelHandlerGpu::test_moe_gate_ignore_layers`
However, the shared fixture currently creates the tiny model as follows:
```python
tiny_model_path = save_tiny_model(
"Qwen/Qwen1.5-MoE-A2.7B",
"./tmp/tiny_qwen_moe_model_path",
num_layers=2,
)
Contributor guide
Assessment
This issue has not been assessed yet.