New converted model files
#3
by
whitphx
HF Staff
- opened
No description provided.
Strange, the weights are much larger in these new conversions. This points to an issue with the weight deduplication step, most likely. Can you confirm which versions of optimum, onnxslim, onnx, and transformers you are using?
I am able to reproduce with the latest versions; can you open an issue in https://github.com/huggingface/optimum?
Will close this PR once,
and create another one only with the newly added quantized versions with proper conversion.
whitphx
changed pull request status to
closed