bddppq
df2e1d2962
Disallow using the OOP api workspace as context managers ( #6456 )
2018-04-09 22:13:54 -07:00
Lu Fang
aab0bd3c13
Change onnx_optimizer API ( #6290 )
2018-04-06 13:46:53 -07:00
Andrew Tulloch
0df662c67f
[Caffe2] [Int8] More exhaustive unit tests for int8 ops (+ bug fix in Int8Add in-place case)
...
As title. This catches one bug in the Int8Add in-place case,
which wasn't tested in int8_test.cc
2018-03-30 21:00:44 -07:00
James Reed
47a1fd208f
Quick and dirty raw value substitution from zip file ( #2454 )
2018-03-29 19:18:58 -07:00
James Cross
16312e8123
[fbtranslate/onnx] decoder step (pytorch -> caffe2) exporter for fbtranlsate
...
This code introduces a new class for exporting decoder step (ensemble) models trained with fbtranslate pytorch to Caffe2 models via ONNX, for the purpose of use in "component beam search" being developed concurrently in C++ by @juancarabina.
2018-03-27 18:10:39 -07:00
Orion Reblitz-Richardson
1d5780d42c
Remove Apache headers from source.
...
* LICENSE file contains details, so removing from individual source files.
2018-03-27 13:10:18 -07:00
bddppq
425361af6a
Bump onnx opset version ( #2402 )
2018-03-23 10:48:12 -07:00
bddppq
f3b7b2f293
Remove ONNX consumed_inputs ( #2278 )
...
* Remove ONNX consumed_inputs
* Bump up opset version to 6 issued by onnx caffe2 frontend
2018-03-22 20:24:35 -07:00
Yinghai Lu
45da53f478
Remove Python onnx-caffe2 conversion code ( #2362 )
...
* WIP
* Remove Python onnx-caffe2 onversion code
* Fix build
* Comments
* Add comments
* Fix typo in comments
2018-03-22 11:59:03 -07:00
Joseph Spisak
b2c56eb219
Removed special handling for onnx sqrt ( #2353 )
2018-03-21 21:05:25 -07:00
bddppq
c155842cc1
Update onnx frontend to emit new onnx Reshape (with shape as input) ( #2287 )
...
* Update onnx frontend to emit new onnx Reshape (with shape as input)
* Address comments and revert submodule change
2018-03-16 16:32:35 -07:00
James Reed
e8f14f5d37
Fix ONNX backend for MatMul ( #2273 )
...
* Fix ONNX backend for MatMul
* Update Python implementation
* Address comments
2018-03-15 14:43:52 -07:00
Yinghai Lu
7e6693991d
Onnx caffe2 backend ( #2039 )
...
* C++ version of ONNX->Caffe2 backend
* use namespace ONNX_NAMESPACE
* Fix Build
* Comments
* Change namespace from onnx_caffe2 to caffe2::onnx
2018-03-12 15:18:05 -07:00
James Reed
60299e03cf
Report all errors during ONNX backend translation rather than failing fast ( #2210 )
2018-03-09 10:58:22 -08:00
Lu Fang
52460a0b30
Add outputs_info as parameter in run_node ( #2161 )
2018-03-09 10:44:51 -08:00
anderspapitto
c0304c83b1
Copy some outputs in order to decouple storage ( #2105 )
...
so that mutating one of them does not mutate the others
2018-03-01 13:25:31 -08:00
anderspapitto
749a17661c
Introduce padding op to mimic pytorch semantics in ONNX export ( #2069 )
...
In pytorch, after pad_packed_sequence, the "extra" elements (after the
ends of the sequences) are reset. In the equivalent Caffe2 graph
exported via ONNX, they contained some leftover values, which caused
tests to fail. Probably no one depends on these values, but just in
case, set them to zero to mimic pytorch semantics.
2018-02-28 15:44:54 -08:00
anderspapitto
76304300a8
Transpose shape inference ( #2057 )
...
* fix variable name
* enhance shape inference to handle transpose
in the case arising from pack_padded(..., batch_first=True)
2018-02-27 11:51:10 -08:00
anderspapitto
ec547ce640
RNN ONNX export: concat hidden/cell states on the right axis ( #2055 )
...
Test Plan: existing tests in onnx-fb-universe catch this, modulo a bug
in the tests which I am fixing in a separate diff
2018-02-26 11:04:04 -08:00
anderspapitto
38f18c1daa
add third output in onnx -> caffe2 lstm conversion ( #2011 )
2018-02-22 17:43:33 -08:00
Yinghai Lu
cc7e61c88d
Move onnx-caffe2 inside caffe2 ( #1921 )
...
* Move onnx-caffe2 inside caffe2
* Update to the lastest onnx-caffe2 and update jenkins env
* Rename onnx_caffe2 to onnx
* Add __init__.py to caffe2/python/onnx
* Change CI check variable to JENKINS_URL
* Cherrypick recent onnx-caffe2 update
2018-02-20 13:56:52 -08:00
