Граф коммитов

16117 Коммитов

Автор SHA1 Сообщение Дата
microsoft-github-policy-service[bot] 10a8ffcf50
Microsoft mandatory file (#3870)
Co-authored-by: microsoft-github-policy-service[bot] <77245923+microsoft-github-policy-service[bot]@users.noreply.github.com>
2022-09-23 16:06:50 +02:00
liqunfu e939648002
disable fp16 test for poolWithSequenceAxis (#3810) 2020-03-31 08:55:14 -07:00
liqunfu e1467a79dc
Merge pull request #3806 from microsoft/liqun/wrap_maxlpool_with_reshape
support Pooling ops with Sequence axis
2020-03-27 14:42:04 -07:00
liqun fu a2055f6e3d fix a typo 2020-03-27 14:40:05 -07:00
liqun fu 80e9f79ae6 update with reviewers' comments 2020-03-27 14:35:31 -07:00
liqun fu 9a7dd4c82b add comments 2020-03-25 12:37:46 -07:00
liqun fu 280ec145e0 support Pooling ops with Sequence axis 2020-03-25 11:20:15 -07:00
Thiago Crepaldi c7bc93fde7 Fix Windows Test Dockerfile 2019-08-06 08:30:35 -07:00
Thiago Crepaldi 9a4bbfe30a Update Windows Test Dockerfile
According to https://techcommunity.microsoft.com/t5/Containers/Windows-Server-2019-Now-Available/ba-p/382430,
'latest' tag is no longer supported. Using '1607' (windows server 2016)
2019-07-31 10:49:03 -07:00
Yang Chen bda8fcbca0 address CR 2019-07-22 17:24:42 -07:00
Yang Chen c3d744010b add FutureValue Op's custom attributes into Scan node's description 2019-07-22 16:48:15 -07:00
Yang Chen 91b78a77cc Merge branch 'yanchen/load-from-memory'
Signed-off-by: Yang Chen <yanchen@microsoft.com>
2019-07-20 01:19:03 +00:00
Yang Chen f843f916cc Merge branch 'yanchen/rm-asserts'
Signed-off-by: Yang Chen <yanchen@microsoft.com>
2019-07-19 02:23:45 +00:00
Yang Chen 3ea8507632
Merge pull request #3722 from microsoft/yanchen/output-mapping
added a mapping from onnx outputs to cntk outputs into the graph's description
2019-07-18 15:11:00 -07:00
Yang Chen d01ef8fa6d remove an assert that seems to be too conservative 2019-07-18 10:22:37 -07:00
Yang Chen cfe8f6ca0a Added a new API to load models from memory
* added a new C api CNTK_LoadModel_FromArray that loads a model
  from memory

* also enabled loading ONNX models from buffer
2019-07-18 00:57:45 -07:00
Yang Chen f7f1e8dc98 added a mapping from onnx outputs to cntk outputs into the graph's description 2019-07-18 00:39:00 -07:00
KeDengMS 99a272bf0c Fix hard tabs 2019-07-05 16:18:23 -07:00
KeDengMS a4621a8a5d
Export custom attributes on Times node to ONNX file (#3709) 2019-07-05 13:33:45 -07:00
liqfu 13f3b5fd3c Merge branch 'master' into liqun/scan14 2019-06-27 10:26:40 -07:00
liqfu abf09855eb transpose batch with dir before reshape to [seq, batch, dir*hidden] for RNN ops 2019-06-27 10:25:47 -07:00
Thiago Crepaldi ac7a549cf1 Fix V2LibraryCSTests end to end test 2019-06-25 23:20:25 -07:00
Thiago Crepaldi 9756a11dfb Fix V2LibraryCSTests end to end test 2019-06-25 12:16:56 -07:00
liqfu ef6731746b set shape bacth dim to 0 for reshape op after LSTM to make batch size editable after ONNX export. 2019-06-21 14:50:18 -07:00
liqfu 03a8c414e1 default keep dim is true 2019-06-19 09:24:52 -07:00
liqfu 1fe0bf1aae reduction keep dim edge cases 2019-06-19 09:17:45 -07:00
liqfu ad1709c5e7 fix reduce op keepdim setting 2019-06-18 15:16:03 -07:00
liqfu 248ff78639 onnx exporter for past/future value: fix for offset > sequence length cases; output delayed initial values 2019-05-10 17:03:27 -07:00
liqfu 8ed70502bf ONNX convolution with more than 1 non-feature/channel dimensions 2019-05-01 14:06:26 -07:00
liqfu 22a86bf3ce onnx export: fix cases when scan input needs to be broadcasted 2019-04-29 11:58:00 -07:00
Thiago Crepaldi 0e172db667 Fix build 2019-04-24 11:20:17 -07:00
liqfu 82b7a392fb merge release 2.7 2019-04-19 11:28:14 -07:00
Thiago Crepaldi 6b72436abd Adds CNTK libs folder to LD_LIBRARY_PATH at CNTK init routine on Python 2019-04-09 12:54:42 -07:00
rpengms b8edaaba83
Rpengms/fp16 bs bugfix (#3632)
Update CNTK.core.bs to add support for precision overwrite for every node
2019-04-09 09:25:37 -07:00
rpengms 9688ac6aea
fp16 - brain script add to CNTK.core.bs (#3617) 2019-04-01 10:14:06 -07:00
Thiago Crepaldi 764c8c4e31 Fix CNTK Tutorial 303 2019-03-25 11:08:19 -07:00
Thiago Crepaldi 49d4c2dea7 Update OOBE GPU Dockerfile to CUDA 10 2019-03-24 21:39:38 -07:00
Thiago Crepaldi f48e11c123 Update conda to >= 4.3.30 2019-03-22 15:12:50 -07:00
rpengms 3b7e3b9357
merge fp16_bs fix (#3609)
fp16 brain script fix accumulator
2019-03-22 02:07:51 -07:00
rpengms 4003c087a1
Merge fp16 brainscript work (#3606)
* FP16 BrainScript - address code review comments

* Remove Tab and fix debug build breaks

* Fix Linux Build breaks

* fp16 brain script - add _CRT_SECURE_NO_WARNINGS

* fp16 brain script - fix NetworkTests

* Update tests for model version change

* Remove changes for InputAndParamNodes

* Fix typo

* Remove redundant code

* Fix optional parameters
2019-03-20 11:36:16 -07:00
liqfu 45ae386bc8 disable Manual/Tutorial tests on Windows Python2.7 2019-03-17 14:36:13 -07:00
Peggy Peng 9f60d8f815 Remove tab which caused build breaks 2019-03-15 10:47:16 -07:00
Peggy Peng 19ccd9f746 Address code review comments 2019-03-15 10:47:15 -07:00
Peggy Peng 3a21b0adbb FP16 all reduce. Cast gradients to and from float16 before sending over NCCL to do all reduce. 2019-03-15 10:47:14 -07:00
liqfu 4765f69c7a support onnx export of 2GB models 2019-03-14 09:16:23 -07:00
Peggy Peng 22ded4679b Address code review comments 2019-02-28 14:43:45 -08:00
Peggy Peng d46ad9d7a3 Update 2019-02-28 14:43:45 -08:00
rpengms 773379e622 Add more comments 2019-02-28 14:43:45 -08:00
Peggy Peng d2dc1234c0 Fix BMUF fp16 to update local learner's fp32 cache 2019-02-28 14:43:45 -08:00
Peggy Peng 26e26e4078 Fix BMUF fp16 bug. Local learner ResetSmoothedGradients should NOT wipe cached fp32 model/parameters. 2019-02-28 14:43:44 -08:00