pytorch/xla

Pull Requests Commits

Enable more instant tracking of the samples per second rate.

dlibenzi committed 7 years ago

59b89809

Update the PyTorch commit id to latest

Alex Şuhan committed 7 years ago

fc8a5667

Add client side compitation cache.

dlibenzi committed 7 years ago

01aa6b66

Prepare XRT computations in parallel.

dlibenzi committed 7 years ago

f2bf8c4e

Move InputListAttr to anonymous namespace

Alex Şuhan committed 7 years ago

f8d36d3b

Some code restructuring and naming cleanup.

dlibenzi committed 7 years ago

a4b3d278

Fix convolution to fit new arguments.

dlibenzi committed 7 years ago

4df06aa3

Fix XRTAllocate to not use the now-goge API anymore.

dlibenzi committed 7 years ago

e65b99cd

Fetch new TF bits to unbreak CrossReplicaSum HLO proto changes.

dlibenzi committed 7 years ago

113af200

Build upstream pytorch as part of nightly torch_xla builds

jysohn23 committed 7 years ago

Verified 67459323

Build upstream pytorch as part of nightly torch_xla builds

jysohn23 committed 7 years ago

a50bb35a

Fix several warnings

Alex Şuhan committed 7 years ago

6bd85fa2

Use per session feeds to avoid copies when scaling on multiple hosts.

dlibenzi committed 7 years ago

8b30c9d3

TODO-- : Compile in Parallel.

dlibenzi committed 7 years ago

f11aa237

Update the PyTorch commit id to latest

Alex Şuhan committed 7 years ago

5622d42f

Run LowerAllTuples pass on the forward graph

Alex Şuhan committed 7 years ago

8a5fe46c

Remove some old multi-op code which is no more used.

dlibenzi committed 7 years ago

66ea2b19

Fix RunComputations() to accept many computations even in case of replicated execution.

dlibenzi committed 7 years ago

06dab6ac

Update torch commit version and use .torch_commit_id instead

jysohn23 committed 7 years ago

Verified c6c5ede1

Fix Github commit ID to be 7 characters

jysohn23 committed 7 years ago

ca244f8a

Add the torch-stable wheels bucket directory to build

jysohn23 committed 7 years ago

5e198814

Get pytorch/pytorch commit ID from .torch_commit_id

jysohn23 committed 7 years ago

6ee1a69a

No need to pack source anymore

jysohn23 committed 7 years ago

3111df22

Update builds to point to newer PyTorch version

jysohn23 committed 7 years ago

1a870ab3

Allow iterator regeneration.

dlibenzi committed 7 years ago

c78f7ba8

Add logging to our passes

Alex Şuhan committed 7 years ago

e667d2f6

Change the readme to point to newest PyTorch

Alex Şuhan committed 7 years ago

9101e295

Changed the computation client interface to be compile handle based to get around the continuous marshaling of computation protos.

dlibenzi committed 7 years ago

1de9c055

Allow an XRT cached node to have multiple outputs/operations.

dlibenzi committed 7 years ago

e842b9fa

Added ability to run train session with fake data.

dlibenzi committed 7 years ago

07de6585

Older