Switch branches/tags
2.0rc/breakingChanges Aghagolzadeh-brainscript_useV2Aggregator_true Aghagolzadeh-dockerfile_AzurebatchAI BMUF-float16 BMUF_fp16 CUDA-elementwise-rework Compliance2.2 Dev KaldiReaderOnWindows LastKnownGood MasterBeforeRemovingUnusedCode aayushg/Seq2SeqRepro aayushg/autoencoder aayushg/seq2seqChanges aforesti/compnetwork-constructors alahiri/distrib-reading-LMSequenceReader alexeyk/bn_fix alexeyk/bn alexeyk/conv_fix alexeyk/conv_tests alexeyk/cudnn3 alexeyk/legacy alexeyk/logsum alexeyk/perf alexeyk/tests alexeyo/Docker-Image alexeyo/ReadMe-News-February-10 alexeyo/fix-linux-configure-script alexeyo/fix-linux-install-mpi-path alexeyo/test-winodows-drop-new-file-names alexeyo/1bit-SGD-License-Update-Do-Not-Integrate alrezni/MatrixQuantizerTestsFailRepro alrezni/RandomGeneratorCpu alrezni/RandomUniform alrezni/ReseedCudaGenerator alrezni/bptt_ctf alrezni/checkpointing_v2 alrezni/ctf_nan alrezni/cuda_reset_test alrezni/divering_loss_with_checkpointing alrezni/fix_pytests alrezni/index_cache alrezni/learners_v2_serialization alrezni/learners_v2_uts alrezni/learners_v2 alrezni/mlf_fix alrezni/mpi_lib_loading alrezni/proposal_layer alrezni/reader_randomization_test alrezni/simple_mnist_repro alrezni/text_reader_benchmark alrezni/v2_communicator alrezni/xpacker alrezni/xpacker2 amitaga/apiChangesForRC amitaga/batchNorm amitaga/beta11BugFixes amitaga/bn amitaga/cudaErrorBugFix amitaga/cudaMemCheckForE2ETests amitaga/dataInteropFeatures amitaga/fix1BitPerf amitaga/fixSimpleAggregatorPerf amitaga/htkmlfreaderMemoryOptimizations amitaga/mblayoutMismatchError amitaga/memoryDebug amitaga/noDEBUG amitaga/sparseMemAllocFix amitaga/v2Library arturl-uwp-doc arturl-uwp-sample-pick-model arturl-uwp-sample barlaso/SampledTimes binbzha/backstitch binbzha/fsmn binbzha/latency-control-blstm-2.5 binbzha/latency-control-blstm binbzha/low-frame-rate binbzha/tf-lstm-refine binxu/LeakyReLU bmitra/NDRMReader bmitra/fix boren/bfsmn_binary_mlf_on_release_2.6 boren/bfsmn_binary_mlf boren/bfsmn_binary_mlf_2.6 boren/release_2.6_kax_fsmn bospoort-maartenb/cntk_103a bowbao/arguments_order bowbao/cuda10_stage bowbao/cuda10 bowbao/freedim_perf bowbao/maxunpooling bowbao/mkl_asym_pad_and_channel_pad bowbao/mkl_channel_pad bowbao/onnx_conv_kerneldim bowbao/onnx_flatten bowbao/onnx_gather_prelu bowbao/onnx_init_typetoproto bowbao/onnx_lrn_mvn_layerNorm bowbao/onnx_match_number bowbao/onnx_matmul_gemm bowbao/onnx_models_unittest bowbao/onnx_slice bowbao/onnxrt bowbao/onnxruntime_ci_stage bowbao/onnxruntime_ci bowbao/overhaul_conv_export bowbao/preserve_work bowbao/protobuf_version bowbao/seqconv_perf bowbao/seqconv_profiling bowbao/seqconv_v2_stage bowbao/seqconv_v2 bowbao/seqconv broda/tolower cazhan/lu-seq-reader chanliu/AMTrain chanliu/TSLearn chaojunl/lfmmi chaojunl/smbr chazhang/faster_rcnn chazhang/pretrain chenta/add_keras_learner chenta/align_axis_indices chenta/auto_broadcast_binary_op chenta/crash_investigate chenta/disable_omp chenta/distPerf chenta/expose_gradient_api chenta/fix_conv_crash chenta/fix_crash chenta/fix_gather_bug chenta/fix_nightly_build_crash chenta/fix_slice_free_dimension chenta/free_dimension_pooling chenta/freedimension_test chenta/keras-final chenta/keras_learner chenta/lotus_test chenta/nightly_crash chenta/padding chenta/print_omp_thread chenta/profile_test chenta/reverse_support chenta/splice chenta/test chenta/times chenta/univeral-learner chenta/unpack_batch_grad_fix chenzhehuai/lfbmmi-negstream-sil chenzhehuai/lfbmmi-negstream clemensm/cntkmkl-evalx clemensm/cntkmkl clemensm/dynamicaxis clemensm/fixseqcla clemensm/mblayouts clemensm/perf2 clemensm/perf4b clemensm/perf4 clemensm/profiler clemensm/quantmultiply2 clemensm/runtimeopt clemensm/s2s cntk_agg cntkteam/new_onnx_without_batch_axis cntkteam/onnx_without_batch_axis cntkteam/onnx1.0 cntkteam/onnx1.2 csnugetquickfix dapisani/center_loss dapisani/cntk_converter dongyu/addhypertune dongyu/htkreaderext dongyu/writerfix duli/histogram duli/tensorboard_image duli/tensorboard_image1 duli/tensorboard_image2 duli/tensorboard_staging ebarsoum/cleanup ebarsoum/cntkv2doc ebarsoum/dilation_with_padding ebarsoum/emotion ebarsoum/inceptionv3 ebarsoum/iteration ebarsoum/layerconv ebarsoum/pixelcnn ebarsoum/png16 eldak/adaptiveLearningRate eldak/addingMLFTests eldak/addingPerDimMeanVarDeNormalization eldak/alignedBundler eldak/cesparsebs eldak/chunkDiagnostics eldak/clonePreservingIds eldak/codeGen4 eldak/compositeInMemory eldak/exposingCWrapperForHalide eldak/exposingGPUThruC eldak/exposingSequenceIdFromMinibatchData eldak/fixingBatchNorm eldak/fixingBlockReplace eldak/fixingCrop eldak/fixingMaskForEval eldak/fixingUDInitialization eldak/forJacob eldak/gpuPrefetchGpuTranspose eldak/includingCNTKLibraryC eldak/investigatingBM eldak/jdEvalC eldak/maxNumSequences eldak/muditsChanges eldak/muditsChanges2 eldak/muditsChanges3 eldak/netRefactoring eldak/noRandomizerChunking eldak/opencv32 eldak/optimizeO eldak/optimizingMBSize eldak/optimizingSparseRead eldak/parallelPacking eldak/refactoringForInference eldak/refactoringForInference2 eldak/renamingManual eldak/retryToOpenFile eldak/sequenceKeyLength eldak/tempFixInVectorMax eldak/ud eldak/udRandom eldak/udRandom2 fmegen/eval-internal-test fmegen/eval-vs2015 fmegen/eval-vs2015_2 fmegen/linux-debian fmegen/vs2015/changing-project-files fseide/complexity fseide/cudnn5 fseide/dynamite fseide/e2easr fseide/fixAsMatrix fseide/fixLoad fseide/fixSamples fseide/kdd fseide/ndops fseide/netlib fseide/pathPrefix fseide/refactor_nodes fseide/s2s fseide/s2sd fseide/s2sd5 fseide/s2spy fseide/seqtest fseide/testlstm fseide/99 gaizna/cntk_readerless gaizna/eval_import_lib gaizna/writeminibatch_fix gather_from_target gather_op giwye/lightrnn gloch/fix_native_proposallayerlib gorand/lstm_experiments guoli/EMBR guoye/EMBR guoye/EMBR1 guoye/dcfsmn guoye/dcfsmn1_debug guoye/dcfsmn1 guoye/dcfsmn2 guoye/dcfsmn4 guoye/master_debug guoye/merge_embr hecli/PostReleaseMerge hjl/Char-RNN-Example hunter/fast-rcnn hunter/fast-rcnn2 hunter/fix-write issaca iu_layer iu.triplet_loss ivanst/profiler_chr ivanst/profiler_cr ivanst/profiler_dir ivanst/profiler_ovr ivanst/profiler ivanst/profiler2 ivanst/profiler3 ivanst/profiler5 ivanst/profiler_20160510 ivanst/profiler_20161108 ivanst/temp ivanst/temp2 jacde/CNTKEvalPlainCAPI jacde/PlainCAPIPlaceholder jaliyaek/htk_test jaliyaek/int16 jdroppo/ceiling-mbsize_conttrain jdroppo/ceiling-mbsize jdroppo/cudnn-rnn-lstm-lfmmi jdroppo/cudnn-rnn-lstm jiajia/DropOutParam2 jiajia/IRMetricV2 jiajia/IRMetricV4 jiajia/cdssm jiajia/cdssm2 jiajia/cdssm3 jiajia/loglossfix jiajia/ndcgcost jianblin/CDenseReader jianblin/CDenseReader1 jianluan/test1 jiashua/CSparsePCReader jieli/CTC jieli/LFMMI jlie/CSparseDensePCReader jlie/CSparsePCReader jlie/SparseDensePCReader jqian/fixnccl jqian/logcublas jqian/nonccl jqian/wip_io kate/dockerFile kedeng/boostAsync kedeng/cpuNcclFix kedeng/cuda92 kedeng/fix2963 kedeng/fixAutopad kedeng/fixCrash kedeng/mkldnn_merge kedeng/mkldnn kedeng/mkldnn2 kedeng/ncclUUID kedeng/oldPackThreshold kedeng/perfTest kedeng/profileReader kyiwasak/faster-rcnn-distributed lakshayg-asinh lakshayg-atanh lakshayg-swish lewlu/AM-Softmax lewlu/ResNet-V2 lewlu/a-softmax lewlu/densenet lewlu/facereaderopt lewlu/inceptionv3-py lewlu/msra-face lewlu/optimizer lewlu/python-disableRegInBN lewlu/rmsprop lewlu/ssgd_experiment_wofixedkernel lewlu/subminibatch-test lewlu/update-aspectratio-limit lewu/AM-softmax lingfw/adadelta_bs lingfw/conttrain-ceiling-mbsize lingfw/lfbmmi-negstream lingfw/lfmmi linux-gcc liqun/Broadcast1252970 liqun/CSApiHighLevel liqun/Lang liqun/Misc2.6Stage liqun/Misc2.6Stage2 liqun/Misc2.6 liqun/Mise2.6_2 liqun/NewSequenceSliceStage liqun/NewSequenceSliceStage2Stage liqun/NewSequenceSliceStage2 liqun/OfficeModel2Stage liqun/OfficeModel2Stage2 liqun/OfficeModel2Stage3 liqun/OfficeModel2 liqun/RNN2.6 liqun/SequenceSlice liqun/SequenceSliceStage liqun/TopK liqun/TopKStage liqun/fixp27Stage liqun/fixp27 liqun/gslfix liqun/gslfixstage liqun/hashDev liqun/issues liqun/iter2.6 liqun/liqun/RNN2.6.Stage liqun/memleak liqun/onnx_axis_stage liqun/onnx_axis liqun/onnx.cast.stage liqun/onnx.cast liqun/onnx1.3Stage liqun/onnx1.3Stage2 liqun/onnx1.3 liqun/protobuf.upgrade liqun/scan liqun/scan2Stage liqun/scan2Stage2 liqun/scan2 liqun/scan3Stage liqun/scan3 liqun/scan4Stage liqun/scan4 liqun/scan5 liqun/seqops liqun/seqopsStage liqun/topk2 liqun/2954 liwchang/StagePreorderTraverse4 liwchang/diagTimesBackCompat liwchang/stagingDiagTimesFix lnunes/add-class-activation-map-example lnunes/cpu_conv_engine_improvements lnunes/remove-gpu-nms mahilleb/AsyncThrow mahilleb/BaselineNew mahilleb/CheckGitVsVs mahilleb/CmakeWip mahilleb/Cntkv2DelayLoad mahilleb/DocTest mahilleb/E2eWithTimeTool mahilleb/FixLinInstall2 mahilleb/LinuxOobeJupyterTestDevice mahilleb/MnistDownload mahilleb/PyLibLocations mahilleb/PyLibLocations2 mahilleb/PyScripts mahilleb/SetupPy mahilleb/ShareVersion mahilleb/SphinxUpd2 mahilleb/TestDriver mahilleb/TestDriverIncludeSku mahilleb/TestDriverNvsmiExit mahilleb/TestDriverTimeoutWip mahilleb/UnitTestEnable mahilleb/Win1BitSgdSeparateOutDir mahilleb/disable104 mahilleb/docfxPostprocess mahilleb/eldak/opencv32 mahilleb/test mahilleb/test1 mahilleb/test2 mahilleb/tut104 mahilleb/u16 mahilleb/valgrind malcolm-cuda-changes manikj/nightly manikj/unifyBuildFolders marhamil-fix-jar marhamil/cherry-pick-jar marhamil/fix-jar-test marhamil/jar-2.2 marhamil/jar marhamil/java-bindings-backup marhamil/java-bindings-one-commit master maven-central-jar meaghago mradmila/dgx1 mradmila/modeltime mradmila/nogpu mseltzer/exportdbn_mkl mseltzer/ivectorreader mseltzer/utterancesourcemulti mseltzer/uttreaderfix msrpeng/sparse_agg_v2_direct_port msrpeng/sparse_agg_v2_sandbox msrpeng/sparse_agg_v2_sandbox1 msrpeng/sparse_agg_v2_sandbox2 muditj/FixHTKLatticeRelatedSettings2 muditj/FixNoRandomizerBug muditj/FixNoRandomizerBug2 muditj/FixNoRandomizerBug3 muditj/FixReaderMemoryLeak2 muditj/FixReaderMemoryLeak3 muditj/FixReaderMemoryLeak4 muditj/FixReaderMemoryLeak5 muditj/HTKDeserializerPythonWrapper muditj/V231NoRandDivergenceDebug muditj/fixBMUFTestcase muditj/pythonCRlossAvg_staging muditj/pythonCRlossAvg muditj/pythonCRlossAvg_2 muditj/pythonCRlossAvg_3 n17s-patch-1 naotous/translated-tutorials nikmil/acc_agg_fix_2 nikmil/cv_best nikmil/ndarray_conversion_public nikmil/python_crop_cr nikmil/python_distrib_eval_cr_rebased nikmil/save_outputs_public nikosk/bidaf nikosk/clang nikosk/gmatch nikosk/lrdisplay nikosk/rangan_repro nikosk/tuned_bidaf nvidia/dgx1 origin/t-titan/attention pahua/CDenseReader penhe/issue_repro penhe/native_cosdistance penhe/reasonet_contractive_loss penhe/reasonet_example penhe/reasonet_tutorial penhe/reasonet penhe/rsn_issue_repro peykash/BN_FP16_fix peykash/TimesTranspose_export peykash/batchnorm_and_pooling_fixes peykash/batchnorm_fixes peykash/batchnorm_float16_fix peykash/batchnorm_float16_fixes peykash/batchnorm_float16_support peykash/batchnorm_fp16_fix peykash/float16 peykash/fp16-gru-fix peykash/fp16-load-from-rawdata peykash/fp16-misc-fixes peykash/initializer peykash/onnx_no_batch_axis_fixes peykash/pooling_backward_compatibility_fix peykash/pretrained_model_fixes peykash/python27_test_fix peykash/reduce_fixes peykash/reduction_allaxes_export peykash/sequence_isfirst_islast peykash/softmax_fix peykash/times_export_fix pkranen/arbDim pkranen/bnFrozenFix pkranen/csevalFastrcnn pkranen/hierClass pkranen/hierarachical_FastRCNN pkranen/mbSize pkranen/new_yolo pkranen/newHierarchy pkranen/odScripts pkranen/yolo prototypes/arm64-debian pytest/update-anaconda-and-pytest qiwye/asgd-dev ratsur/find-by-uid ratsur/jar-rc3 ratsur/java-bindings ratsur/java-csharp-parity-rebase ratsur/java-csharp-parity ratsur/java-training ratsur/maven-central-jar release/latest release/2.0.beta2.0 release/2.0.beta3.0 release/2.0.beta4.0 release/2.0.beta5.0 release/2.0.beta6.0 release/2.0.beta7.0 release/2.0.beta8.0 release/2.0.beta9.0 release/2.0.beta10.0 release/2.0.beta11.0 release/2.0.beta12.0 release/2.0.beta15.0 release/2.0.rc1 release/2.0.rc2 release/2.0.rc3 release/2.0 release/2.1 release/2.2 release/2.3 release/2.3.1 release/2.4 release/2.5 release/2.5.1 release/2.6 removeSNB resetbranch rpengms/HTK_more_logging rpengms/fp16_bmuf rpengms/fp16_bmuf_20 rpengms/fp16_sandbox ruizhao/CTC_HC ruizhao/CTC_NDL ruizhao/CTC_Nbest ruizhao/CTC_blankinmlf ruizhao/CTC_checkin2 ruizhao/CTC ruizhao/LFMMI ruizhao/RNNT_V2 ruizhao/RNNT_bs ruizhao/RNNT ruizhao/RNNT_830merge ruizhao/SE_TS ruizhao/SE ruizhao/SMBR ruizhao/base ruizhao/lfmmi-2pass-m15 ruizhao/lfmmi-2pass-test ruizhao/lfmmi-2pass ruzhao/CTC_leftTC sanjeevm/test_branch sayanpa/aqgan sayanpa/aqganchk sayanpa/contribmd sayanpa/gi2022 sayanpa/nbsphinx5 sayanpa/nbsphinx6 sayanpa/seqclass sayanpa/seqclass2 sayanpa/slawek sayanpa/sr sayanpa/srcin sayanpa/srcin2 sayanpa/srcin3 sayanpa/srcin4 sayanpa/t303chk sayanpa/t303 sedymche/fix-doc sedymche/message sedymche/onnx-backend-test sedymche/onnx-backend sedymche/onnx-broadcast sedymche/onnx-maxes-new sedymche/onnx-maxes sedymche/onnx-min-max sedymche/onnx-reshape sedymche/onnx-save-load sedymche/test-skip shuzhe/treeps singletonComm slayton58/divmod sparse_agg sparse_aggregation_large_BUCKETSIZE sparse_aggregation_simple_selectK sparse_aggregation sparse_cntk sptiwari/KEEP_onnx_remove_autopad_from_conv sptiwari/UpdateFasterRcnn sptiwari/add_zeros_ones_eye sptiwari/add_zeros_ones_eye2 sptiwari/conv_backend_test_fix sptiwari/conv_master_failure sptiwari/conv_with_static_batch_axis sptiwari/convtranspose_update sptiwari/convtranspose_update2 sptiwari/convtranspose_update3 sptiwari/convtranspose_update5 sptiwari/convtranspose_update6 sptiwari/convtranspose_update7 sptiwari/dilated_conv_non_mkldnn_test sptiwari/dilated_conv sptiwari/disable_test_conv_setup sptiwari/elu_fix2 sptiwari/fasterRcnn_NoFastMath sptiwari/fasterRcnn_freeDim3 sptiwari/group_conv_cudnn7 sptiwari/group_conv_input_validation sptiwari/mkldnn_first_round sptiwari/mkldnn_first_round2 sptiwari/mkldnn sptiwari/mnist_onnx_group_conv_fix sptiwari/model_fix_test sptiwari/one_hot_onnx sptiwari/one_hot_onnx2 sptiwari/ones_zeros_like_op sptiwari/onnx_avg_pool_stride sptiwari/onnx_conv_reload sptiwari/onnx_conv_reload2 sptiwari/onnx_convtrans_fix2 sptiwari/onnx_convtranspose_fixes sptiwari/onnx_emotion_fix sptiwari/onnx_emotion_fix1 sptiwari/onnx_emotion_fix2 sptiwari/onnx_emotion_fix3 sptiwari/onnx_hash_update sptiwari/onnx_hash_update2 sptiwari/onnx_inception_fix sptiwari/onnx_inception_fix2 sptiwari/onnx_infra_for_simple_batch_ops2 sptiwari/onnx_lrn_fix sptiwari/onnx_merge_with_master sptiwari/onnx_pooling_overhaul sptiwari/onnx_resnet3d sptiwari/onnx_topk sptiwari/onnx1.2_merge sptiwari/shufflenet sptiwari/straight_through_op sptiwari/straight_through_op2 sptiwari/test_onnx_branch_reversion sptiwari/update_depth2space sptiwari/update_onnx_metadata szhao/cntkport t-alkhar/cntk-board t-alkhar/distributed-sequence-example t-alkhar/distributed-testing t-alkhar/graph-output-bug-fix t-alkhar/resnet-example-fix t-alkhar/tensorboard-cntk-connect t-alkhar/visualization-tool t-anqli/gan t-ivrodr/adding_more_cpp_cs_eval_examples t-ivrodr/binary_reader_changes_rev t-ivrodr/ctfparser_multithread t-ivrodr/fix_for_bufferedreader t-ivrodr/fix_percentage_epoch_progress t-ivrodr/fix_segfaul_nonprimitive_func t-ivrodr/naming_nodes_resnet_example t-ivrodr/refactoring_readers_file_handling t-ivrodr/setting_py_value_to_deserializer t-ivrodr/small_fixes_to_training_log t-ivrodr/uncommenting_cs_examples t-ivrodr/using_filewrapper_in_bin_reader t-kax/bfsmn t-kax/debug-fsmn t-kax/fix-param-share-bug t-kax/fsmn t-makrei/YOLOv2 t-makrei/hierClass t-makrei/hierarchical_FastRCNN t-makrei/new_yolo t-makrei/transpose-fix t-mofun/learners_grad_clip_doc t-mofun/rmspropgraves t-tiah/LMNCE-dev t-tiah/lm t-tiah/reader-debug t-tidett/ExtendedCUDAErrors t-tidett/memoryswapping t-tidett/memoryswappingWithFlagToTrue t-tidett/onDemandMemorySwapping t-titan/attention t-titan/shiftread thhoens/NewBinaryReader thhoens/atomic-nosync-nop thhoens/atomic-nosync thhoens/atomic thhoens/basemat thhoens/binreader thhoens/bnfix thhoens/cdssm thhoens/cuda7 thhoens/distgradfix thhoens/mblayout thhoens/multi thhoens/multivalid thhoens/nbr thhoens/nbr2 thhoens/newbinreader thhoens/nosync thhoens/packmatrix thhoens/pathfix thhoens/perf2 thhoens/perf4 thhoens/reader thhoens/readertest thhoens/s2s thhoens/seq2seq thhoens/sparsematrix thhoens/sparsemul thhoens/syncguard thhoens/test thhoens/unatomic-nosync-nop thhoens/unatomic-nosync thhoens/unatomic thiagofc/clean thiagofc/ctf_parser_perf thiagofc/ctf_parser_test thiagofc/disable-csharp-target-for-linux thiagofc/disable-failing-tests thiagofc/disable-many-tests thiagofc/fix-cntk-dev-docker thiagofc/fix-debug-target-on-linux thiagofc/fix-docker-image thiagofc/increase-batch-norm-spatial-error-tolerance thiagofc/make-master-green thiagofc/revert-mkl-dnn-upgrade thiagofc/temporarily-disable-doctests thiagofc/test-reset thiagofc/update-anaconda-and-pytest thiagofc/update-mkldnn-to-0.14-on-windows thiagofc/update-mkldnn-to-v0.14 tix/FixParallelBuild tix/convertEvalExampleTestToNetCore tix/fixExamples tix/memoryPinning tix/testCifar users/fmegen/brainwave-cntk-beta15 v-igquin/fix-binary-conv v-igquin/fix-doc-forward-backward v-lelu/facereader-test v-lelu/facereader v-lelu/inception-resnet-v2 v-lelu/inceptionv4 v-niveli/mem_sharing_supressed v-niveli/mpi_bcast_with_profiler v-rodemo/disable-node-timing v-rodemo/fix-batchnorm-freedimension v-yuxgu/caffe2cntk v-yuxgu/crosstalkcaffe-bugfix v-yuxgu/model2cntk v-yuxgu/psroipooling v-yuxgu/rpn v-yuxgu/sigmoid v-zhke/fb-ssgd v-zhke/model2cntk_testcase v-zhke/smb_verify v-zhke/ssgd_experiment vad vadimma/AddLogs vadimma/AddSaveLogs vadimma/BMUF vadimma/CTC_R vadimma/CTCDebug vadimma/CTCNode vadimma/EMBR vadimma/EMBR1 vadimma/Eval_optimizations vadimma/ExtendedEvalExample vadimma/FixLN vadimma/FixRepetitiveValueUpdate vadimma/LatticeDebug vadimma/LatticeParallel vadimma/ModelVersionLog vadimma/MomDouble vadimma/MomentumLog vadimma/QuantizedParameter vadimma/ReaderPrint vadimma/RemoveCodeDuplication vadimma/SE_debug vadimma/adambs vadimma/badlat vadimma/binaryMLF vadimma/binaryMLF2 vadimma/binaryMLF3 vadimma/binmlf vadimma/ctc_ndl vadimma/evaluateStreamMode vadimma/exposeSE vadimma/fb vadimma/fixCNTKCompat vadimma/fse vadimma/latency-control-blstm-2.5 vadimma/latpar vadimma/lc-blstm-sq vadimma/maxerrors vadimma/mlfreader vadimma/nccl_bmuf vadimma/reader_c vadimma/se_de vadimma/sed vadimma/stableTanh vadimma/testfp16 vadimma/testlr vadimma/wu vadimma/wuf vadimmia/binary vlivan/large-minibatches vlivan/linux-gpu-locking vlivan/linux-reader-catchup vlivan/linux-reader-catchup2 vlivan/test-master wdarling/axisdoc wdarling/lstmaux wdarling/pythongru wdarling/tutorial wdarling/update_fastrcnn wdarling/updatezeroes weba/FixBatchNorm weixi/autolr weixi/conttrain weixi/lfmmi-2-lstm weixi/lfmmi-2pass-cr weixi/lfmmi-2pass-test weixi/lfmmi-2pass weixi/lfmmi-conttrain-neg weixi/lfmmi-conttrain-wlstm weixi/lfmmi-conttrain weixi/lfmmi-debug weixi/lfmmi-memzip weixi/lfmmi-new weixi/lfmmi weixi/oldbuild weixi/pcv weixi/waynecoding wilrich/cntkv2Cython wilrich/cntkv2Swig wilrich/rl wolfma/checkdep wolfma/instsplat wolfma/mkl2017 wolfma/setup wolfma/t-makrei/transpose-fix xiaoyongzhu-master xwang/tts-lace yacheo/h-softmax-fix yacheo/h-softmax yanchen/copy-api-internals yanchen/null-output-owner yanchen/primitive-func yanchen/wchar yaronwe/internal_protobuf yaronwe/internalprotobuf yingsh/OrderedCEWithSM yuqing/HDF5Reader yuqtang/AugmentMinibatchSourceAPI yuqtang/ChangeOneHotDefaultToSparse yuqtang/DebugCrossTalkConv2d yuqtang/DebugGPUMemory yuqtang/DyanmicChunkDescriptions yuqtang/EnableDefMBSizeInStreamInfo yuqtang/FixBugsInNoBackPropOpsGradients yuqtang/FixDistributedTrainingCounts yuqtang/FreeAxisTensorTimes yuqtang/FunctionalLearningParameter yuqtang/GradOnGatherIndices yuqtang/ImplDefMbSizeStreamForUserDeserializer yuqtang/LambdaDeserializer yuqtang/LightWeightParameterTracker yuqtang/RecurrenceStepFunctionSignatureCheck yuqtang/SetDiagonalOnNonSquaredMatrix yuqtang/StageDebugClone yuqtang/StageDefMbSizeUserDeserializer yuqtang/StageEyeLike yuqtang/StageFixCrossTalkConv2d yuqtang/StageGradOnGatherIndices yuqtang/StageNoBackPropOPsGraidentFixes-backup yuqtang/StageNoBackPropOPsGraidentFixes yuqtang/TimesOnFreeAxes yuqtang/ValidateRecurerenceStepFunction yuqtang/add-hierarchical-names-to-layers yuqtang/debugClone yuqtang/debugDeserializer yuqtang/eye_like yuqtang/query-answer-embed yuqtang/query-answer-embed2 yuqtang/query-answer-embed3 yuqtang/refactorReaderChunkInfo zhal/randomSeedOffset zhouwang/compositereader-determinismus zhouwang/cseval-unittest zhouwang/eval-pythonmodel zhouwang/fix-v2localstatic zhouwang/fix-value-readonly-master zhouwang/fix-value-readonly zhouwang/github1921 zhouwang/keran-issue zhouwang/kevinpan-memoryissue zhouwang/renbo-issue zhouwang/tlc zhouwang/valueptr-issue zhouwang/valueptr-issue2 zhujie/SpDsRowStack_Maxpooling zhujie/SpDsRowStack zhujie/SparseRowStack zhujie/binary_step_continuation_pretrain zhujie/binary_step_continuation_quant zhujie/binary_step_continuation zhujie/binary_step_continuation_01 zhujie/cdssm2_binary zhujie/deterministic-st-tanh zhujie/elementmax-nwords-3bit-srs zhujie/elementmax-nwords-annealtanh zhujie/elementmax-nwords-srs_v2 zhujie/elementmax-nwords-srs zhujie/elementmax-nwords-st1 zhujie/elementmax-nwords zhujie/elementmax zhujie/matmul-baseline-readerfix zhujie/matmul-baseline zhujie/matmul-cusparse zhujie/sewt-densebase zhujie/sewt zhujie/sparseandx zhujie/sparsemulandx zhujie/sparserowstack zhujie/stochastic-st-dropout zhujie/stochastic-st-signp1 zhujie/stochastic-st-tanh zhujie/stochastic-st zhujie/straightthrough-check zhujie/straightthrough_tensorops zhujie/straightthrough_unaryop zhujie/straightthrough zhujie/treenode-cleaf zhujie/treenode zhujie/vecmaxpooling zsoltm/profiler
Nothing to show
Find file History
Permalink
Type Name Latest commit message Commit time
..
Failed to load latest commit information.
data
README.md
char_rnn.py

README.md

Build Neural Character Language Models with CNTK

This example demonstrates how to build a neural character language model with CNTK using regular plaintext data fed in using the numpy interface.

A neural language model uses a recurrent neural network to predict words (or characters) with a richer context than traditional n-gram models allow. In this implementation, a character is run through an LSTM and the output is then put through a fully-connected layer to predict the next output character. The model can learn to be extremely expressive as the context is progressively built-up with each letter run through the RNN. For even more expressiveness, we allow a stack of LSTMs where the output of each layer is put through the next layer as its input.

This example is inspired by Andrej Karpathy's blog post "The Unreasonable Effectiveness of Recurrent Neural Networks" at http://karpathy.github.io/2015/05/21/rnn-effectiveness/ and his accompanying code at https://github.com/karpathy/char-rnn. This example allows you to achieve similar results to those displayed in Karpathy's blog, but with the packed-sequence training efficiency that CNTK allows.

HOWTO

Set the hidden_dim and num_layers to values that match the complexity/size of your data. To learn a model, simply call train_lm(your_data) where your_data is a plaintext file containing your training data. Once you have a model that you're happy with (the code is currently configured to save a new model at the end of each epoch [i.e. each pass over the full training data]), then call the function load_and_sample as in the following example:

load_and_sample("models/shakespeare_epoch19.dnn", "tinyshakespeare.txt.vocab", prime_text=text, use_hardmax=False, length=100, temperature=0.95)

In the above, we pass in the model saved in models/shakespeare_epoch19.dnn (i.e. the saved model after training for 20 epochs), the vocab tinyshakespeare.txt.vocab (which is automatically created from tinyshakespeare.txt when you train a model with that training data), the prime-text text (which will run some priming text through the model before sampling from it), use_hardmax set to False meaning that there will be some sampling instead of just always taking the most likely predicting from the model, the length of the sample you wish to generate (in characters, including the prime-text), and finally, the temperature where 1.0 means use the actual probabilities predicted by the model, and lower numbers flatten the distribution so that the samples will be less like the learned model but more "creative".

Have fun!

Examples

Using the tiny-Shakespeare data (included) with a 2-layer LSTM and hidden-dim of 256:

KING up:
low to it; for he's mistress that I might see,
I spurn them in good words.

BEVIS:
Then sport what!
Madam, the vein o' the ill highest of the hide.

KING JOHN:
Fie, Henry, if thou be my reverend courage,
Whose two spurs poorer in or partless,
Yet riveted by his eld, il execution,
Lukess undout sound teach; four wives do sworn,
As with this carf's--God woo, to this! what tends?
Till this unlewn bushes are but fourteen,
Or sitter on our pyn; and on my better
A drum and fitness of my bearing

Using the 20-newsgroup data with a 2-layer LSTM and hidden-dim of 256:

Newsgroups: ut.whuhroel
Date: 20 Apr 93 15:03:08 GMT
Lines: 19


> I am annoying.  This way to the tapes of principle about everyone like validity must do you.

Dick. Good chunkan

mjson@austin.ibm.com
>-->

Xref: cantaloupe.srv.cs.cmu.edu comp.sys.ibm.pc.hardware:61655 alt.sockha.mangers.36784558 talk.religion.misc:99703
comp.os.ms-windows.misc:9720 comp.aiz.dack-orit:15@stp.noscne.com>
Organization: IBR, Aucidore.
Reply-To: jjlu@llo@minta.UU.NOT
>>Adutions a road-Tell 9:20 AX   10516 PHKRS

>BA