pytorch/xla

Pull Requests Commits

Define explicit entrypoint for container

jysohn23 committed 6 years ago

fc83941c

Trivial fixes over #831

dlibenzi committed 6 years ago

c9b69ddc

fix avg_poo2d avg_pool3d

Ailing Zhang committed 6 years ago

86462216

Added divisor_override to avg_pool2d and avg_pool3d for compatibility with pytorch

ifedan committed 6 years ago

21c85220

Fixed clang-format-7 indent.

dlibenzi committed 6 years ago

35d8e530

Make tests pickup clang-7 by default (#832)

jysohn23 committed 6 years ago

Verified 1f4b1128

The BN backward might miss running_{mean,var} data, and since we are not using it ATM, drop it from passing it downstream.

dlibenzi committed 6 years ago

31d200cd

Install wheels within conda environment (#829)

jysohn23 committed 6 years ago

Verified 69b4cbe4

Install googleapiclient and oauth2client in conda (#826)

jysohn23 committed 6 years ago

Verified ce532453

Rename guide file as this will contain more then simple "perf" things.

dlibenzi committed 6 years ago

f8f8918d

Compare integer tensors for equality instead of using AllClose

asuhan committed 6 years ago

93b973db

Add TPU performance guideline document

asuhan committed 6 years ago

f6413081

Update README.md

jysohn23 committed 6 years ago

Verified 351dac09

Wait for the client mesh master to be up before moving on.

dlibenzi committed 6 years ago

232d7f28

Fix last use of old Status based MultiWait API.

dlibenzi committed 6 years ago

9856d1dc

Make MultiWait to use exceptions instead of Status.

dlibenzi committed 6 years ago

2b89bd1a

Make client mesh master initialization parallel.

dlibenzi committed 6 years ago

0e7cb393

Use clang-7 compiler in cpp tests (#816)

jysohn23 committed 6 years ago

Verified f5012f84

Explicitly cast to float when doing tensor compares.

dlibenzi committed 6 years ago

e160b400

Fix batch normalization to be able to deal with train() and eval() modes.

dlibenzi committed 6 years ago

257ddd13

Fix C++ tests warnings.

dlibenzi committed 6 years ago

c1ae6801

Allow the loop functions to save device specific data (ie, optimizers, in case of them carrying LR schedules) across training loop.

dlibenzi committed 6 years ago

b1f944ac

Merge pull request #804 from ailzhang/refactor_cpp_test

ailzhang committed 6 years ago

Verified ef4ff9ff

Ailing Zhang committed 6 years ago

52ba56f7

Merge branch 'master' of https://github.com/pytorch/xla into refactor_cpp_test

Ailing Zhang committed 6 years ago

49abdbc0

use oss config for clang-format

Ailing Zhang committed 6 years ago

c0f348d4

search & replace with a wrapper to .to(..., /*non_blocking=*/, /*copy=*/)

Ailing Zhang committed 6 years ago

32e7863b

use AtenDefaultDevice

Ailing Zhang committed 6 years ago

87a9f6aa

ForEachTorchXLADevice->ForEachDevice

Ailing Zhang committed 6 years ago

5b53bfd3

refactor cpp tests

Ailing Zhang committed 6 years ago

44a7c3c6

Older