junpeiz

Coremltools has added `glu` op

1

Just FYI, the blocker (`glu` op) of exporting [Speech2Text](https://huggingface.co/docs/transformers/main/en/model_doc/speech_to_text) model has been added into coremltools in [this PR](https://github.com/apple/coremltools/pull/1614).

Will you release the code?

10

Hi, thank you for your release of the system output, but is there any plan to release the training and testing code?

The `embedding_4bit` implementation has a strong assumption about weight_quant_min and weight_quant_max

2

### 🐛 Describe the bug In the `embedding_4bit` implementation [here](https://github.com/pytorch/executorch/blob/main/exir/passes/_quant_patterns_and_replacements.py#L213), it assumes the quantized data has `quant_min=-8`, so it uses `weight = weight.view(torch.int8).add(-8)` to shift the data. However, the shifting...