fxmarty

Sorry to pollute this issue then, I will post this single message. Yes, spresense SDK's `tools/config.py` (e.g. called as `./tools/config.py examples/hello`) would fail for me in `do_olddefconfig()`, when we try...

Any reproducible benchmark on this?

@pommedeterresautee Maybe you have some insights on this? I read https://github.com/ELS-RD/transformer-deploy/blob/d397869e95ee07570c47edefec01bdc673391b65/docs/faq.md#why-dont-you-support-gpu-quantization-on-onnx-runtime-instead-of-tensorrt , but it's not clear to me why ONNX Runtime + `TensorrtExecutionProvider` would be worse than Tensor RT native,...

That's great! We should include the new subclasses in https://github.com/huggingface/evaluate/blob/main/docs/source/package_reference/evaluator_classes.mdx I think, and probably mention them in the how-to guide.

It is not clear why there is such a limit.

Great! In any case, even if `CUDAExecutionProvider` is solely passed, the execution providers from onnxruntime is still a list, `["CUDAExecutionProvider", "CPUExecutionProvider"]`, so to me it makes still sense to have...

Will be really useful to try out graph modifications on top of quantization I think!

It looks nice, more flexible than onnxruntime! The following code ```python from transformers import AutoModelForImageClassification from torch.ao.quantization import get_default_qconfig from optimum.fx.quantization import quantize import torch model_name = "google/vit-base-patch16-224" model =...

Also, I could not get the `quantize()` to work for static quantization on vision tasks, related to what appears to me to be a bug in datasets (but I could...

An other weird behavior I came upon: ```python from transformers import AutoModelForImageClassification, AutoFeatureExtractor from transformers.utils.fx import symbolic_trace from torch.ao.quantization import get_default_qconfig from optimum.fx.quantization import quantize from datasets import load_dataset model_name...