🚀 Feature request

At the moment we cannot return a list of attention weight outputs in Flax as we can do in PyTorch.

In PyTorch, there is a output_attentions boolean in the forward call of every function, see here which when set to True collects

We keep this issue open to collect feature requests from users and hear your voice. Our monthly release plan is also available here.

You can either:

1. Suggest a new feature by leaving a comment.
2. Vote for a feature request with 👍 or be against with 👎. (Remember that developers are busy and cannot respond to all feature requests, so vote for your most favorable one!)
3. Tell us that

🚀 Feature

Detect UninitializedParameter and run one batch/sample before fitting.

Motivation

Pytorch now accepts 'lazy' layers with UninitializedParameter.

However, this seems to cause a memory error in PL at when we start the trainer because it attempt to estimate the memory usage:

RuntimeError: Can't access the shape of an uninitialized parameter. This error usually happen

Change tensor.data to tensor.detach() due to
pytorch/pytorch#6990 (comment)
tensor.detach() is more robust than tensor.data.

Bug Report

Is the issue related to model conversion? No

Describe the bug

DynamicQuantizeLinear function op does not have shape inference function defined. In absence of shape inference, function body is used to get the shape inference for the function op and although it works as a fallback option it hurts perf.

Expected behavior

Add shape inference function for DynamicQuan

Is your feature request related to a problem? Please describe.
I typically used compressed datasets (e.g. gzipped) to save disk space. This works fine with AllenNLP during training because I can write my dataset reader to load the compressed data. However, the predict command opens the file and reads lines for the Predictor. This fails when it tries to load data from my compressed files.

I'm using mxnet to do some work, but there is nothing when I search the mxnet trial and example.

Hi
Wikiann dataset needs to have "spans" columns, which is necessary to be able to use this dataset, but this column is missing from huggingface datasets, could you please have a look? thank you @lhoestq