Hi, I've moved to Cuda 11.0 and I am getting warnings below while compiling. I just can't remember exactly, but I don't think I saw these with 10.2. I don't know if it related, but run_test.py is also failing when it comes to Distribution related tests. I can post test's log when the build is complete, if it is helpful. Thank you.

[3950/5005] Building NVCC (Device) object caffe2/CMakeF

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

I just ran into an issue when trying to use to_csv with distributed workers that don't share a file system. I shouldn't have been surprised that writing to a local file system from a distributed worker doesn't work. It shouldn't work. But the error I got was just a File Not Found error. That brought me to:dask/dask#2656 (comment) - which was the answer.

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.

Description

Move the adamw optimizer defined here to MXNet. https://github.com/dmlc/gluon-nlp/blob/20321598a732d254d4a9e036eecbe0ae8abb48a0/src/gluonnlp/optimizer.py#L31-L32

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), ["