It should be added to the .rst so that they appear on the website.
Doc is already inline here https://github.com/pytorch/pytorch/blob/df88cc3f7f5c13221a93d7d3d38e681a3d5a6b6a/torch/nn/modules/module.py#L89-L115

cc @jlin27 @mruberry

Improve navigation of the I/O how-to by prepending a decision tree/flowchart using an image or graphviz, as suggested by @mattip.

PR #6447 adds a public API to get the maximum number of registers per thread (numba.cuda.Dispatcher.get_regs_per_thread()). There are other attributes that might be nice to provide - shared memory per block, local memory per thread, const memory usage, maximum block size.

These are all available in the FuncAttr named tuple: https://github.com/numba/numba/blob/master/numba/cuda/cudadrv/drive

Is there a way to train a bidirectional RNN (like LSTM or GRU) on trax nowadays?

The current version of the HANS dataset is missing the additional information provided for each example, including the sentence parses, heuristic and subcase.

• What's wrong?

Description

We should use the official mxnet batchify functions to implement our own batchify functions. However, since we'd like to later support other frameworks, we should still keep our own batchify.py. We can change it to call MXNet implementations.

MXNet batchify: https://github.com/apache/incubator-mxnet/blob/master/python/mxnet/gluon/data/batchify.py
GluonNLP batchify: https://gi

Support Series.median()

MCVE Code Sample

# Your code here
import numpy as np
import xarray as xr

data = np.zeros((10, 4))
example_xr = xr.DataArray(data, coords=[range(10), ["