pytorch/xla

Pull Requests Commits

Add benchmark for tensor send to backend.

dlibenzi committed 6 years ago

677f3252

Drop diplicated import and reorder.

dlibenzi committed 6 years ago

5af84c34

Avoid repeating adding the optimizer's state.

dlibenzi committed 6 years ago

3abdf335

Make sure the copy partition size is never 0. This could have happened, for example, in a machine with many cores (ie, > 64), and a tensor of shape (32, 32, 32, 32, 32).

dlibenzi committed 6 years ago

9fe779a7

Added option to save HLO test on file, in case of error, as log lines gets intermingled.

dlibenzi committed 6 years ago

2755abd3

Dump the XLA HLO graphs in case of compile errors.

dlibenzi committed 6 years ago

f1f57f31

Do not print the loss at every step.

dlibenzi committed 6 years ago

328ccddf

Fix threaded copy code to avoid selecting the strided copy dimension as threading dimension.

dlibenzi committed 6 years ago

3addbe8f

Slice a tensor copy over multiple threads.

dlibenzi committed 6 years ago

0b1e034d

Tune hyperparams for ResNet18 on cifar

jysohn23 committed 6 years ago

56d53b5d

Allow hyper parameter tuning and scale LR for MNIST multi core. (#682)

dlibenzi committed 6 years ago

69c50145

Script to update torch and torch_xla installation (#681)

jysohn23 committed 6 years ago

Verified 0848d793

Fix build breakage due to API changes.

dlibenzi committed 6 years ago

37363b46

Fix PyTorch Tensor virtual function signature change.

dlibenzi committed 6 years ago

698eec31

Collect torch build artifact

jysohn23 committed 6 years ago

Verified 04353988

Add support for max and min across dimensions

asuhan committed 6 years ago

7cb7ba04

Merge pull request #674 from pytorch/jysohn/openblas

jysohn23 committed 6 years ago

Verified 352db72e

Install OpenBLAS as part of wheel build setup

jysohn23 committed 6 years ago

e70059fe

Merge pull request #673 from pytorch/jysohn/pyversion

jysohn23 committed 6 years ago

Verified 87064fb1

Explicitly use python3 for test/run_tests.sh

jysohn23 committed 6 years ago

c98ec35e

Limit the check graph frequency as well, until the XLA bug has been fixed.

dlibenzi committed 6 years ago

769109e4

Use common syn function for apply tensors graph.

dlibenzi committed 6 years ago

0a99322c

Add support for _th_le, _th_ne

asuhan committed 6 years ago

72caf51f

Send more tensors in parallel to allow the parallelism used by the XRT computation client to reduce the send time.

dlibenzi committed 6 years ago

1723eb55

Cache the Tensorflow tensor memory allocations to reduce clear_page_c performance hits.

dlibenzi committed 6 years ago

2360dae0

Remove more unused code in test/test_operations.py

asuhan committed 6 years ago

d8899009

Added simple thread pool instead of spawning new threads.

dlibenzi committed 6 years ago

d581df31

Fix scalar type of sum

asuhan committed 6 years ago

8885ba67

Fix the hanging issue on multi-core.

dlibenzi committed 6 years ago

92c08e06

MultiCore API tweaks and adding back removed models, now using the MultiCore API.

dlibenzi committed 6 years ago

ad330b33

Older