Commit Graph

4896 Commits

Author SHA1 Message Date
Marcin Junczys-Dowmunt
8ecc8b653f FindMKL modified to find installed mkl in system paths 2021-02-24 05:19:34 +00:00
Roman Grundkiewicz
f88ded2ba8
Add documentation platform based on Sphinx+Doxygen+Breathe+Exhale (#803)
* Setup Sphinx+Doxygen+Breathe+Exhale
* Remove links to modindex and search pages
* Enable Doxygen autobrief
2021-02-23 16:25:30 +00:00
delong-coder
927b95a3ab
Update marian-backend (#786)
* Update marian-backend
* Fix bug
2021-02-22 13:26:55 +00:00
Roman Grundkiewicz
6627134064
Update VERSION 2021-02-22 13:01:38 +00:00
Rihards Krišlauks
b29c50f9b7
Update simple-websocket-server to the latest version (#799)
This adds support for boost 1.75
2021-02-22 13:01:03 +00:00
Marcin Junczys-Dowmunt
147d9dc25b do not do dropout at inference 2021-02-12 05:06:39 +00:00
Marcin Junczys-Dowmunt
93cdfdcc9a enable marian-tgz building for non-static build, fix small dropout bug 2021-02-12 04:39:13 +00:00
Roman Grundkiewicz
80b74b8441 Merged PR 17572: Add Azure pipeline with make install
A new Azure pipeline testing `make install`.
2021-02-08 16:09:28 +00:00
Martin Junczys-Dowmunt
5aeea4e066 Merged PR 17430: Refactors MPI interfaces and adds different types of gradient exchanges
* Refactors MPI-related code
* Adds node-local updates with occasional inter-node updates
* decouples batch-reading across nodes
2021-02-08 05:27:49 +00:00
Marcin Junczys-Dowmunt
24e2e01027 merge with internal master 2021-02-07 00:24:03 +00:00
Marcin Junczys-Dowmunt
6f6d484665 increase version to 1.10.0 2021-02-06 15:35:16 -08:00
Roman Grundkiewicz
bd9feacf49 Merged PR 17548: Fix 'no disk space left' in Azure Pipelines jobs
Removing `build/src` after compilation seems to fix an issue with 'no disk space left' in Ubuntu 18.04 CUDA 10.2 gcc-8 pipeline that we see recently.
2021-02-05 19:15:50 +00:00
Marcin Junczys-Dowmunt
916a790241 Merge branch 'master' of vs-ssh.visualstudio.com:v3/machinetranslation/Marian/marian-dev 2021-02-04 09:13:49 -08:00
Marcin Junczys-Dowmunt
ff89b0ae13 set transformer-depth-scaling to false by default to avoid backwards compat issues 2021-02-04 09:13:40 -08:00
Young Jin Kim
4894b303b9 Merged PR 17540: Code clean-up for avx512 support
Made the code cleaner and added comments
2021-02-04 07:05:32 +00:00
Young Jin Kim
fd60221ace Merged PR 17503: AVX512 file generation in prod
Added avx512 file generation
2021-02-04 06:11:21 +00:00
Young Jin Kim
b2d2a5d457 Merged PR 17525: cmake fix for GENERATE_MARIAN_INSTALL_TARGETS
cmake fix for GENERATE_MARIAN_INSTALL_TARGETS
2021-02-04 05:04:48 +00:00
Roman Grundkiewicz
e407d49bda Merged PR 17467: Enable pipeline with CUDA 11.2
Enabling Azure job with Ubuntu 20.04, CUDA 11.1, gcc-9, static, GPU-only.
2021-02-01 19:40:46 +00:00
Martin Junczys-Dowmunt
ba91b391e2 Merged PR 17337: fp16 support for training
This PR refactors the training graph groups and optimizers to enable and simplify things for fp16 support.

Deprecates old unused graph groups and fixes a couple of MPI issues.
2021-01-28 16:15:44 +00:00
Roman Grundkiewicz
d490d461d2 Update regression-tests 2021-01-26 02:40:03 -08:00
Roman Grundkiewicz
aa1b6fd134 Update regression-tests 2021-01-25 06:42:45 -08:00
Roman Grundkiewicz
024de9a4ad
Update VERSION 2021-01-25 14:39:51 +00:00
Roman Grundkiewicz
587b8a8403
Fix intgemm merge (#793) 2021-01-25 12:09:34 +00:00
Marcin Junczys-Dowmunt
18fd50df85 Bump up version 2021-01-24 16:03:49 -08:00
Marcin Junczys-Dowmunt
d29e41af25 Merge branch 'pmaster' 2021-01-24 16:03:11 -08:00
Nikolay Bogoychev
600f5cbdec
Integrate intgemm into marian (#595)
Adds intgemm as a module for Marian. Intgemm is @kpu 's 8/16 bit gemm library with support for architectures from SSE2 to AVX512VNNI
Removes outdated integer code, related to the --optimize option

Co-authored-by: Kenneth Heafield <github@kheafield.com>
Co-authored-by: Kenneth Heafield <kpu@users.noreply.github.com>
Co-authored-by: Ulrich Germann <ugermann@inf.ed.ac.uk>
Co-authored-by: Marcin Junczys-Dowmunt <marcinjd@microsoft.com>
Co-authored-by: Roman Grundkiewicz <rgrundkiewicz@gmail.com>
2021-01-24 16:02:30 -08:00
Qianqian Zhu
737f43014a
Fix to resolve run time failures for FastOpt enabled WASM builds (#779)
* copy changes from commit 4df92f2
* add comments for better understanding
* restore the newline at the end of file and add this changes in changelog.md
2021-01-07 13:12:36 +00:00
Roman Grundkiewicz
c1c4af08a9
Bump version 2021-01-07 10:42:24 +00:00
Tommy MacWilliam
f59008276c
Add support for Apple Accelerate (#769)
* support for Apple Accelerate
* add a CMake flag to use Apple Accelerate as the BLAS library.
* rename USE_ACCELERATE to USE_APPLE_ACCELERATE
* add comment with more info on Accelerate
* link to the Apple documentation on Accelerate.
2021-01-07 10:41:55 +00:00
Roman Grundkiewicz
542be2f909
Bump version 2021-01-06 11:30:58 +00:00
Graeme
49994a0460
Prefer if/else in place of try/catch in FastOpt makeScalar (#774) 2021-01-06 11:30:22 +00:00
Roman Grundkiewicz
3b468e4628
Remove ::set-env from GitHub checks for Windows (#766) 2020-11-19 20:12:28 +00:00
Marcin Junczys-Dowmunt
ced5b0faf3 Merge branch 'pmaster' 2020-11-14 17:59:35 -08:00
Roman Grundkiewicz
2637a454ad
Update Boost in GitHub checks to 1.72 (#764) 2020-11-13 11:24:00 -08:00
Roman Grundkiewicz
a6436668cc Merged PR 16429: Use Boost 1.72 in Azure pipelines
Update to Boost 1.72 and explicitly use the ubuntu-18.04 image.
2020-11-13 19:17:25 +00:00
Marcin Junczys-Dowmunt
c8d4457bec Merge branch 'master' of github.com:marian-nmt/marian-dev into pmaster 2020-11-13 10:40:52 -08:00
Marcin Junczys-Dowmunt
69dc82c360 Fixes #763 2020-11-13 10:40:03 -08:00
afaji
5990142047
Fix the allocation size for opt-step Tensor in the quantizer (#761) 2020-11-11 12:37:28 +00:00
Marcin Junczys-Dowmunt
a7268a36db update version based on number of PRs from last time 2020-11-10 18:24:38 -08:00
Marcin Junczys-Dowmunt
99e06610a2 merge with internal master 2020-11-10 16:39:47 -08:00
Martin Junczys-Dowmunt
9dad84ae9b Merged PR 16337: Update sentencepiece to new version
This updates the SentencePiece version in Marian to a much more recent revision. Due to that there is no dependency on Protobuf anymore.
2020-11-11 00:38:37 +00:00
Martin Junczys-Dowmunt
b90229d8ee Merged PR 16294: Change stopping criterion for mini-batch-fit binary search
This PR changes the stopping criterion for mini-batch-fit binary search to better maximize batch size.
2020-11-10 17:08:47 +00:00
Marcin Junczys-Dowmunt
79b35afe3b update regression test pointer 2020-11-10 07:32:14 -08:00
Marcin Junczys-Dowmunt
bbdccd1c92
Update sentencepiece to newest version (#753)
- Updates sentencepiece to the newest version (removes dependency on protobuf)
- Enable SentencePiece compilation by default since there is no dependency in protobuf anymore.
2020-11-10 07:01:34 -08:00
afaji
65ea504ec8
Hotfix: crash when --quantize-optimization-steps is set to 0 (#759)
* fix problem if the optimization step is set to 0
* set the first error residual to 0
2020-11-09 22:41:55 -08:00
Martin Junczys-Dowmunt
d5e773f937 Merged PR 15925: Training embedder separation with margin losses
* This PR adds training of embedding spaces with better separation based on https://arxiv.org/abs/2007.01852
* We can now train with in-batch negative examples or a handful of hand-constructed negative examples provided in a tsv-file.
2020-11-07 17:46:39 +00:00
Marcin Junczys-Dowmunt
3d233ec592 Merge branch 'master' into pmaster 2020-11-04 14:37:25 -08:00
delong-coder
ca7a887aa7
Add Triton Marian backend running in AzureML Inference Environment (#749)
* Add Triton Marian backend running in AzureML Inference Environment
2020-11-04 14:29:36 -08:00
afaji
e274ac76b2
Quantized model finetuning (#690)
* enable quantized training
2020-11-04 14:25:40 -08:00
Martin Junczys-Dowmunt
f3e4cbf705 Merged PR 16219: Allow to set epoch display width
Allow to set the display width of the fractional pert of a logical epoch.
2020-11-04 17:31:59 +00:00