https://github.com/Microsoft/CNTK

sort by:
Revision Author Date Message Commit Date
7bda219 Add arch=compute_30 codegen for debug builds required by CNTK native BatchNorm implementation 15 February 2016, 21:03:31 UTC
11f11b1 Add printMetadata config to documentation for the dumpNode top-level command. 15 February 2016, 01:23:37 UTC
f89e544 Integrate AddPrintMetadataFlag into master 14 February 2016, 17:31:17 UTC
e1093b9 Integrate f0d8a23b26cbfe94640aa9869dcf57e4a4352455 into master 13 February 2016, 15:28:20 UTC
d2b84a5 Integrate mahilleb/WindowsCpuOnlyTest into master 13 February 2016, 13:54:40 UTC
6ff885a Fixed build warning. 12 February 2016, 19:43:40 UTC
19b0c82 Addressed code review feedback. 12 February 2016, 19:34:03 UTC
9d6f68f Enable Windows CPU-only end-to-end tests (empty commit) 12 February 2016, 19:33:14 UTC
fe997db Fix for printf warning. 12 February 2016, 17:34:31 UTC
03e8cc5 Addressed code review feedback. 12 February 2016, 17:34:17 UTC
e12427d Fix for name resolution. 12 February 2016, 17:34:02 UTC
e8ab337 Refactored Makefile to include defines both to NVCC and g++. 12 February 2016, 17:33:47 UTC
e25e548 Added SM30 config to MathCuda Debug build defaults. 12 February 2016, 17:33:32 UTC
958f634 Fixed builds. 12 February 2016, 17:33:18 UTC
820cfc2 Fixed bug with BN engine param name in BS and fixed CPU build. 12 February 2016, 17:33:03 UTC
11fe2c6 Added eps and engine parameters to BN node. 12 February 2016, 17:32:49 UTC
9ad6147 Batch norm: added epsilon as parameter, enabled run mean/invstddev. 12 February 2016, 17:32:35 UTC
1379f58 Added convolutional implementation for batch norm. 12 February 2016, 17:32:21 UTC
5846fae Improved BN unit tests. 12 February 2016, 17:32:07 UTC
21575f3 Added batch norm per-activation backprop implementation. 12 February 2016, 17:31:53 UTC
4e9483b Added batch norm per-activation feed forward implementation. 12 February 2016, 17:31:39 UTC
6465c80 Integrate mahilleb/02_Convolution_Unstable into master 12 February 2016, 16:14:42 UTC
4e8c3e7 Tests/EndToEndTests/Examples/Image/MNIST/02_Convolution: unstable on Windows 12 February 2016, 15:44:58 UTC
7d24810 Integrate mahilleb/TestDriverWinCpuonly into master 12 February 2016, 15:01:12 UTC
f0d8a23 Add experimental support for OpenBLAS library * Add 'openblas' as mathlib option in configure. Not added to auto-search so must be specified using --with-openblas * configure script searches empty tail so that libraries located at default_path_list roots (ie /usr/local/ + include/openblas_config.h) are found * Treat ACML as the odd library out in ifdefs since it doesn't conform to typical BLAS standard. Other libraries like ATLAS should be able to share OpenBLAS/MKL variants. Add default USE_ACML define in VS projects to match * Fix 'max' macro define colliding with C++ std::max once openblas headers are included Usage Notes: * For best performance, build OpenBLAS with USE_OPENMP=1. When running CNTK, set OPENBLAS_NUM_THREADS environment var or set numCPUThreads CNTK config variable to the physical core count or performance will suffer * OpenBLAS 2.16 (git HEAD) tested in Linux with GCC 4.8.4 and in Windows with OpenBLAS 2.15 (pre-built binary release + MingGW 64-bit support dlls) * For Windows, in Math.vcxproj, replace libacml_mp_dll.lib with libopenblas.dll.a and change USE_ACML define to USE_OPENBLAS. Change ACML_PATH environment variable to your OpenBLAS path. Modify openblas_config.h as per https://github.com/xianyi/OpenBLAS/issues/708 * On current generation Intel processors, OpenBLAS measures a little faster than AMD ACML and slower than Intel MKL on MNIST and other examples 11 February 2016, 21:40:18 UTC
9386b8d Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/htkmlfreaderMemoryOptimizations 11 February 2016, 18:50:57 UTC
e511c9d Addressed CR feedback 11 February 2016, 18:50:49 UTC
7652ef0 Merge pull request #90 from felixonmars/patch-1 Fix a typo: lanugage -> language 11 February 2016, 15:17:21 UTC
c360717 Integrate mahilleb/Msbuild2 into master 11 February 2016, 13:27:19 UTC
416577f Fixing the AV for the partial minibatch 11 February 2016, 09:54:25 UTC
c4ba87d Integrate cd4808e into master 11 February 2016, 08:41:16 UTC
58c768b Temporarily retained old memory inefficient frame randomization for side by side testing. The old randomization is off by default and can be switched on using a config setting minimizeReaderMemoryFootprint=false 10 February 2016, 23:26:58 UTC
3021ad0 Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/htkmlfreaderMemoryOptimizations 10 February 2016, 19:53:23 UTC
4a3b6ea Tests/TestDriver.py: pick up Windows CPU-only binaries from x64/{Debug,Release}_CpuOnly 10 February 2016, 16:52:49 UTC
73b27ef Address CR comments 10 February 2016, 11:18:24 UTC
1b29d2f .vcxproj: let intermediate output go into a common directory Fixes #12. 10 February 2016, 08:10:07 UTC
2c2a721 CNTK.Cpp.props: define properties to distinguish build flavors and use them 10 February 2016, 08:05:31 UTC
ea55623 Introduce CNTK.Cpp.props and load in every .vcxproj 10 February 2016, 08:05:31 UTC
52aad32 .vcxproj: drop include path $(VCInstallDir)atlmfc\include 10 February 2016, 08:05:31 UTC
9c80220 .vcxproj: remove HTKMLFReader IncludePath where not needed 10 February 2016, 08:05:31 UTC
3b979f8 .vcxproj: coalesce importing for .user.props for Debug / Release 10 February 2016, 08:05:31 UTC
c2ccd66 .vcxproj: remove empty <Command/> 10 February 2016, 08:05:31 UTC
2b22008 .vcxproj: remove empty custom build steps 10 February 2016, 08:05:31 UTC
8ffd7e7 CNTK.vcxproj: fix: add missing parameter for prebuild.bat for Release builds 10 February 2016, 08:05:31 UTC
7baf28a Fixed linux build 10 February 2016, 06:18:35 UTC
2ec9d7a Replace incorrect use of bad_exception with runtime_error exceptions 10 February 2016, 05:29:54 UTC
409327a Write GPUMatrix to file in chunked manner to avoid creating very large intermediate CPU buffers 09 February 2016, 21:09:18 UTC
439707a Added new handler to print call stack on allocation failure 09 February 2016, 18:55:09 UTC
57fd382 Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/htkmlfreaderMemoryOptimizations 09 February 2016, 05:43:09 UTC
5945c39 Removed fsync call from fflushOrDie since its is causing issues when writing philly models to HDFS 09 February 2016, 01:34:45 UTC
e9f5ca4 Write checkpoint file before writing model file, so that a checkpoint file is guaranteed to exist for the latest successfully written model file 09 February 2016, 00:12:05 UTC
5a67cf2 Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/htkmlfreaderMemoryOptimizations 08 February 2016, 21:10:49 UTC
c214980 Disabled HTKMLFReader unit tests HTKMLFReaderSimpleDataLoop7 and HTKMLFReaderSimpleDataLoop17 as they use invalid config setting combination of Truncated=true and frameMode=true 08 February 2016, 21:10:40 UTC
a2b99a4 HTKMLFReader memory optimization: Reimplemented frame mode randomization to keep just a window of randomized frames in memory instead of having a full array of all frames in the corpus 08 February 2016, 00:49:40 UTC
cd4808e Added missing quote in case OPENCV_PATH contains spaces. 07 February 2016, 05:46:35 UTC
bb608f8 LSTM test now uses an actual 1D vector for the bias parameter (as opposed to a 1-column 2D matrix) 07 February 2016, 04:31:48 UTC
6887465 fixed a size_t wraparound in TensorSliceWithMBLayoutFor() that broke some 1-column tensors; also sorted m_allRoots by name for better reproducability 07 February 2016, 04:05:57 UTC
f321625 Added a check for inconsistent frameMode and Truncated settings. Also addresed some minor CR feedback. 06 February 2016, 07:06:15 UTC
9279662 Matrix::Print() can now print corners for easier debugging 06 February 2016, 01:10:53 UTC
2f9a48c InputNodes() now skips inputs that are only reachable through PreComputeNodes that have already been computed, addressing Issue #65; cleaned up some unnecessary NULL checks before delete 05 February 2016, 19:06:20 UTC
2bbee29 Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/fixInputNodes 05 February 2016, 17:18:59 UTC
7c86176 temporary fix for "computeGradient" parameter inconsistency 05 February 2016, 17:18:45 UTC
84e9db1 Added a workaround for the HDFS FUSE implementation bug of failure to delete destination file when the destination file specified to rename function already exists 05 February 2016, 02:29:31 UTC
83e7783 better error messages for Win32 exceptions 04 February 2016, 23:42:08 UTC
d3e636c Fixed redundant writing of the .0 model file when no PreCompute is performed and the model is loaded from a checkpoint 04 February 2016, 23:23:40 UTC
9cf21b7 Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/fixInputNodes 04 February 2016, 22:52:02 UTC
98a723c reduced clutter of validation output by validating the entire network, rather than each output node individually 04 February 2016, 22:51:53 UTC
a930d1c Updated EvalActions to fix support for OutputNodeNames in the config. 04 February 2016, 21:11:13 UTC
14881a2 Added guards to all model saving calls to prevent mutiple parallel workers from trying to write the model file concurrently 04 February 2016, 18:55:28 UTC
1e65a41 fixed bug in catching when MPI runtime is missing 04 February 2016, 16:51:08 UTC
17ea344 Fix a typo: lanugage -> language 04 February 2016, 14:20:50 UTC
8b2c9ff Integrate mahilleb/ImageReaderOptDepFix into master 04 February 2016, 12:22:24 UTC
29dfe2d Integrate eldak/addressingReaderCR into master 04 February 2016, 11:28:46 UTC
7914c7c Integrate pkranen/fixMNIST into master 04 February 2016, 11:13:07 UTC
5fa84b0 ImageReader.vcxproj: don't build if OpenCV not present 04 February 2016, 10:00:26 UTC
f5a349a Addressing ImageReader CR from Frank 04 February 2016, 08:37:14 UTC
7d1fc36 fixed .config to .cntk in CSEvalCient 04 February 2016, 08:11:06 UTC
576028b fixed 01.. config to match wiki page 04 February 2016, 08:11:04 UTC
27fd4e0 Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/htkmlfreaderMemoryOptimizations 04 February 2016, 07:05:16 UTC
eb231c3 (added a comment) 04 February 2016, 00:52:47 UTC
0900bdb gave up back-compat with some mis-spelled names of training criterion nodes 03 February 2016, 23:27:00 UTC
78587f5 Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/htkmlfreaderMemoryOptimizations 03 February 2016, 20:19:54 UTC
070006d addressed an issue discovered in static code analysis 03 February 2016, 19:49:13 UTC
0626c5d Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/fixCrossProcessMutex 03 February 2016, 19:36:39 UTC
2d11980 addressed bugs discovered by static code analysis, Issue #81, and manually merged two related pull requests 03 February 2016, 19:32:13 UTC
498e022 Fixed MNIST batch norm sample. 03 February 2016, 18:01:50 UTC
9f9b681 Added README.md with disclaimer to ResNet samples. 03 February 2016, 18:01:49 UTC
c394833 Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/fixCrossProcessMutex 03 February 2016, 17:32:47 UTC
9b1f342 Integrate mahilleb/WindowsVsDependenciesDoc into master 03 February 2016, 17:30:22 UTC
87acb14 final simplification of case-insensitive string comparisons 03 February 2016, 17:24:09 UTC
901f99f further simplified case-insensitive string compares 03 February 2016, 17:17:58 UTC
16175a1 cleanedup all case-insensitive comparisons 03 February 2016, 17:01:55 UTC
f454962 fixed process-name comparison in BestGpu::QueryNvmlData() 03 February 2016, 16:38:24 UTC
7ab3d9e .vcxproj: Let diagnostics about dependencies point to CNTK Wiki 03 February 2016, 16:22:13 UTC
ee97506 Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/fixCrossProcessMutex 03 February 2016, 16:09:37 UTC
ab7d5f5 Integrate mahilleb/BuildAndTestTuneNew into master 03 February 2016, 15:02:45 UTC
fe75bb1 Extend build information included / reported by CNTK * Build target: GPU or CPU-only * Build type (Debug or Release), was missing for Windows * Whether or not 1bit-SGD was included * If there were local modifications in the Git working copy * Generate BUILD{ER,MACHINE,PATH} on Linux 03 February 2016, 14:48:39 UTC
0e7c663 Tools/build-and-test: fixes and extensions * Add --test-targets options to limit test to CPU / GPU CPU builds support testing only for CPU, GPU builds support testing for both. * Add --clean-build-after, hook-up and fix existing code. * Change build directories on Linux to build/$TARGET/$FLAVOR * Fix clean-up bug between test run. * More quoting for 'rm' * Support CPU-only build for Windows * MSbuild in parallel and with less output * Exit if both GPU and 1bit-SGD is requested on Windows (side-by-side not yet support) * GPU tests will specify device ID 0, not Auto 03 February 2016, 14:39:14 UTC
0e5ad89 Reapply fix in Makefile for SparsePCReader Reapply name changes in Makefile that had been resolved but got lost during excessive rebasing. 03 February 2016, 09:15:34 UTC
d23eecc Fixed TestDriver.py output format regressions introduced in dc1ed4628f 03 February 2016, 08:28:17 UTC
back to top