(Draft)

New Features

TODO: Summarize and highlight all new features

New Operators

• split_v2 (#13687)
• Gradient multiplier (contrib) operator (#13632)
• Image normalize operator - GPU support, 3D/4D inputs (#13802)
• Image ToTensor operator - GPU support, 3D/4D inputs (#13837)
• Add Gluon Transformer Crop (#14259)
• GELU (#14449)
• AdamW operator (Fixing Weight Decay Regularization in Adam) (#13728)

Feature Improvements

Operators

• make ROIAlign support position-sensitive pooling (#13088)
• Add erfinv operator for calculating inverse error function (#13811)
• Added optional parameters to BilinearResize2D to do relative scaling (#13985)
• MXNET-1295 Adding integer index support to Sequence* family of operators. (#13880)
• Export resize and support batch size (#14014)
• CUDNN dropout (#13896)
• Relaxing type requirements for slice_like op (#14097)
• Relaxing type requirements for reshape_like op (#14325)
• Parallelize CPU version and add GPU version of boolean_mask op (#14090)
• Add NHWC layout support to Pooling (cpu, gpu cuda, gpu cuDNN) (#13749)
• Multi-precision AdamW update op (#14171)
• [op] add back support for scalar type rescale_grad argument for adamw_update/mp_adamw_update (#14221)
• move choose_element_0index to operator (#14273)
• Optimize NMS (#14290)
• Optimize NMS part 2 (#14352)
• add backgroud class in box_nms (#14058)
• Use cudnn for dropout by default (#14278)
• In-place updates for Nadam, Adadelta, Adamax and SGLD (#13960)
• Aggregate SGD (#13346)
• Add proper exception message for negative shape in array creation routines (#14362)
• Support multi-threading for Custom Operator (#14363)
• moveaxis operator now accepts negative indices and sequence of ints as well. (#14321)
• Support SyncBatchNorm5D (#14542)
• Add nd.power and sym.pow (#14606)
• Change RNN OP to stateful (#14476)
• Add imresize and copyMakeBorder to mx.image (#13357)

MKLDNN

• Feature/mkldnn static (#13628)
• Feature/mkldnn static 2 (#13503)
• support mkl log when dtype is fp32 or fp64 (#13150)
• Add reshape op supported by MKL-DNN (#12980)
• Move the debug output message into MXNET_MKLDNN_DEBUG (#13662)
• Integrate MKLDNN Conv1d and support 3d layout (#13530)
• Making MKL-DNN default on MXNet master (#13681)
• Add mkldnn OP for slice (#13730)
• mkldnn s8 conv API change for master (#13903)
• [MKLDNN] Enable signed int8 support for convolution. (#13697)
• add mkldnn softmax_output (#13699)
• MKLDNN based Quantized FullyConnected Operator and its fusion (#14128)
• Fix entropy for uint8 (#14150)
• Update MKL-DNN to v0.18 release (was: fix the Dense layer issue) (#13668)
• [MKL-DNN] Enable s8 support for inner product and 3d input with flatten=false (#14466)
• Optimize transpose operator with MKL-DNN (#14545)

ONNX

• ONNX export: Instance normalization, Shape (#12920)
• ONNX export: Logical operators (#12852)
• ONNX import/export: Size (#13112)
• ONNX export: Add Flatten before Gemm (#13356)
• ONNX import/export: Add missing tests, ONNX export: LogSoftMax (#13654)
• ONNX import: Hardmax (#13717)
• [MXNET-898] ONNX import/export: Sample_multinomial, ONNX export: GlobalLpPool, LpPool (#13500)
• ONNX ops: norm exported and lpnormalization imported (#13806)
• [MXNET-880] ONNX export: Random uniform, Random normal, MaxRoiPool (#13676)
• ONNX export: Add Crop, Deconvolution and fix the default stride of Pooling to 1 (#12399)
• onnx export ops (#13821)
• ONNX export: broadcast_to, tile ops (#13981)
• ONNX export: Support equal length splits (#14121)

TensorRT

• [MXNET-1252][1 of 2] Decouple NNVM to ONNX from NNVM to TenosrRT conversion (#13659)
• [MXNET-703] Update to TensorRT 5, ONNX IR 3. Fix inference bugs. (#13310)
• [MXNET-703] Minor refactor of TensorRT code (#13311)

FP16 Support

• Update mshadow to support batch_dot with fp16. (#13716)
• float32 → float16 cast consistency across implementations (#13857)
• modifying SyncBN doc for FP16 use case (#14041)
• support dot(vector, vector) for fp16 inputs on GPU (#14102)
• softmax for fp16 with fp32 accumulator (#14098)
• [MXNET-1327] Allow RNN Layers to be initialized to fp16 (#14219)
• fp16 safe norm operator (#14616)

DGL support

• Add graph_compact operator. (#13436)
• Accelerate DGL csr neighbor sampling (#13588)

Horovod Integration

• Add extra header file to export for error checking (#13795)
• whitelist symbols for using MXNet error handling externally (#13812)
• Use CPUPinned context in ImageRecordIOParser2 (#13980)
• Add pin_device_id option to Gluon DataLoader (#14136)

Dynamic Shape

• [MXNET-1315] Add checks for dynamic-shaped operators in CachedOp (#14018)
• [MXNET-1325] Make InferShapeAttr a standalone pass (#14193)
• [MXNET-1324] Add NaiveRunGraph to imperative utils (#14192)

Backend Engine

• Add infer_type_partial (#14214)
• Tidy up storage allocation and deallocation (#14480)
• Add MXEnginePushAsync and MXEnginePushSync C APIs (#14615)
• Enhance subgraph API (#14113)
• Enhance PartitionGraph (#14277)

Numpy

• [MXNET-1330] Bring nnvm::Tuple to mxnet::Tuple (#14270)
• [numpy] Support zero-dim and zero-size tensors in MXNet (#14661)

Large Tensor Support

• Large array support for randint (#14242)
• [MXNET-1185] Support large array in several operators (part 1) (#13418)

Quantization

• Exclude concat layer for gpu quantization (#14060)
• Enhance gpu quantization (#14094)
• Register fake grad to subgraph and quantized operators (#14275)
• Add int8 data loader (#14123)

CoreML

• Add more support for mxnet_to_coreml (#14222)

Website

• Version switching user experience improvements (#13921)

Front End API

Gluon

• Add pixelshuffle layers (#13571)
• [MXNET-766] add dynamic_unroll RNN for HybridBlock (#11948)
• add pos_weight for SigmoidBinaryCrossEntropyLoss (#13612)
• Rewrite dataloader with process pool, improves responsiveness and reliability (#13447)
• Complimentary gluon DataLoader improvements (#13606)

Python

• Python BucketingModule bind() with grad_req = 'add' (#13984)
• Refine runtime feature discovery python API and add documentation to … (#14130)
• Runtime feature detection (#13549)
• Add dtype visualization to plot_network (#14066)
• [MXNET-1359] Adds a multiclass-MCC metric derived from Pearson (#14461)
• support long for mx.random.seed (#14314)
• Optimization of metric evaluation (#13471)

Language Bindings

Scala

• [MXNET-1260] Float64 DType computation support in Scala/Java (#13678)
• [MXNET-1000] get Ndarray real value and form it from a NDArray (#12690)
• Now passing DType of Label downstream to Label's DataDesc object (#14038)
• Scala interpreter instructions (#14169)
• Add default parameters for Scala NDArray.arange (#13816)
• [MXNET-1287] Up scala comp (#14667)

Java

• [MXNET-1180] Java Image API (#13807)
• [MXNET-1285] Draw bounding box with Scala/Java Image API (#14474)
• Add BERT QA Scala/Java example (#14592)

C++

• print error message for mxnet::cpp::Operator::Invoke when failed (#14318)

Clojure

• [Clojure] - Add Spec Validations to the Optimizer namespace (#13499)
• [Clojure] Add Spec Validations for the Random namespace (#13523)
• [Clojure] Correct the versions in the README so they correspond to the latest maven.org release ([#13507)
• Port of scala infer package to clojure (#13595)
• Clojure example for fixed label-width captcha recognition (#13769)
• Update project.clj file to use the snapshots repo to be able to pull (#13935)
• [Clojure] Add resource scope to clojure package (#13993)
• [clojure-package] improve docstrings in image.clj (#14307)
• [Clojure] Helper function for n-dim vector to ndarray (#14305)
• [clojure]: add comp-metric based on CompositeEvalMetric (#14553)
• [Clojure] enhance draw bounding box (#14567)
• [Clojure] Add methods based on NDArrayAPI/SymbolAPI (#14195)
• [Clojure] Clojure BERT QA example (#14691)
• [clojure-package][wip] add ->nd-vec function in ndarray.clj (#14308)

Julia

• Julia v0.7/1.0 support and drop v0.6 support (#12845)
• Julia: split ndarray.jl into several snippets (#14001)
• Julia: split symbolic-node.jl into several snippets (#14024)
• Julia: rename mx.clip to clamp for NDArray (#14027)

Perl:

• Two more gluon loss classes. (#14194)

R

• add NAG optimizer to r api (#14023)
• R-Package Makefile (#14068)

Performance Improvements

• Less cudaGet/SetDevice calls in Gluon execution (#13764)
• Improve bulking in Gluon (#13890)
• Increase perfomance of BulkAppend and BulkFlush (#14067)
• Performance improvement in ToTensor GPU Kernel (#14099)
• Performance improvement in Normalize GPU Kernel (#14139)
• Bulked op segments to allow Variable nodes (#14200)
• Performance improving for MKL-DNN Quantized FullyConnected (#14528)
• speedup SequenceMask on GPU (#14445)
• Dual stream cudnn Convolution backward() with MXNET_GPU_WORKER_NSTREAMS=2. (#14006)
• Speedup _contrib_index_copy (#14359)
• use mkl sparse matrix to improve performance (#14492)

Build and Test

• compatibility with opencv4 (#14313)
• Flaky test #14189 (#14190)
• Enforce determinism for backwards compatibility checker (#14463)
• Change CUB submodule to track Nvidia CUB project. (#13322)
• Updates gpu tests to use CUDNN_VERSION supplied by the environment but default to 7.0.3 if not set (#14595)
• [Dependency Update] Upgrade the libtiff to 4.0.10 (#14623)

Example and Tutorials

• [MXNET-949] Module API to Gluon API tutorial (#12542)
• Support SSD f32/int8 evaluation on COCO dataset (#14646)

Bug-fixes

• Fallback to dense version for grad(reshape), grad(expand_dims) (#13599)
• Fix the bug of BidirectionalCell (#13575)
• set _scale in Trainer using optimizer rescale_grad (#14593)
• [MXNET-1379] update reshape operator (#14600)
• Add repr for SymbolBlock (#14423)
• Cudnn conv dgrad algo filtering (#14310)
• Fix memory leak for size-zero ndarray (#14365)
• Fixes the test_sgld (#14473)
• Revert "Fix memory leak for size-zero ndarray (#14365)" (#14477)
• fix custom operation in fork (#14451)
• Fixes test_operator_gpu.test_multinomial_generator (#14475)
• support leading dimension of -1 in ravel/unravel (#14356)
• begin=end not a valid input (#14403)
• Fix NaN value comparisons in relu, max and min ops (#14262)
• fix engine crash in shutdown phase (#14382)
• fix OOM error during resource allocation (#14444)
• Fix relative difference scala (#14417)
• Correct update count with Gluon trainer and update_on_kvstore=False (#14377)
• Fix crashes on visualization (#14425)
• Reorder module import orders for dist-kvstore (#13742)
• Fixes for trainer with update_on_kvstore=False (#13721)
• Fix errors in docstrings for subgraph op; use code directive (#13463)
• Add resiliency to onnx export code (#13426)
• update github location for sampled_block.py (#13508)
• Revert "Manually track num_max_thread (#12380)" (#13501)
• Revert "Feature/mkldnn static 2 (#13503)" (#13540)
• [MXNET-1110] Add header files required by horovod (#13062)
• [MXAPPS-1020] Clean up some Sphinx warnings. (#13539)
• Fixing a 404 in the ubuntu setup doc (#13542)
• [MXNET-1249] Fix Object Detector Performance with GPU (#13522)
• [MXNET-769] Use MXNET_HOME in a tempdir in windows to prevent access denied due t… (#13531)
• Chi_square_check for discrete distribution fix (#13543)
• Fix use-before-assignment in convert_dot (#13511)
• fix the situation where idx didn't align with rec (#13550)
• fix link for gluon model zoo (#13583)
• Fix exception handling api doc (#13519)
• [MXNET-1253] fix control_flow_op (#13555)
• fix the Float not showing correctly problem (#13617)
• fix quantize pass error when the quantization supported Op are excluded in the model (#13596)
• Fix for import mxnet taking long time if multiple process launched (#13602)
• Revert "Feature/mkldnn static (#13628)" (#13638)
• updated reference to Apache MXNet (#13645)
• Fix incorrect delete in MXExecutorReshape exception handling (#13376)
• add build fix for Scala/Java build (#13655)
• remove omp which can cause ssd accuracy variance (#13622)
• Fix Jetson compilation (#13532)
• Revert "Fix Jetson compilation" (#13665)
• Fix Jetson compilation (#13666)
• Revert "Revert "[MXNET-43] Fix Jetson compilation" (#13665)" (#13672)
• fix unpicklable transform_first on windows (#13686)
• Fix NDArray ToDLPack Bug (#13698)
• Fix the quantization script to support Python2 (#13700)
• Update basic_layers.py (#13732)
• [MXNET-1231] Allow not using Some in the Scala operators (#13619)
• [MXNET-244] Work around likely compiler bug on nested inlines and temporary acces… (#13535)
• Use curl to download sample data instead of wget. (#13761)
• fix bipartite match memory corruption (#13727)
• remove attributs clear on TRT nodes for GetOptimizedSymbol (#13703)
• fix redirection issues; set default version to master (#13796)
• fix for params with no dims in onnx (#13413)
• Remove semicolon in libmxnet.sym file (#13822)
• remove useless code (#13777)
• Fixing a symlink issue with R install (#13708)
• fix minor indentation (#13827)
• Fix Tree Reduction on new instance type p3dn.24xlarge (#13852)
• [Clojure] package infer tweaks (#13864)
• Fix cpp examples build on Mac. (#13826)
• Fix launch bounds in spatial transformer (#13188)
• Update example scripts classpath. (#13849)
• fix ssd quantization script error (#13843)
• Avoid adding SegfaultLogger if process already has sig handler. (#13842)
• fix the fetching GPU problem (#13889)
• Fix SN-GAN example doc (#13877)
• update Spectral Normalization Code (#13868)
• Fixed java benchmark failing error by fixing the classpath (#13891)
• Fix the order of error term's operands (#13745)
• fix bug in nag optimizer (#13683)
• Fix BatchNorm converter for CoreML when fix_gamma=True (#13557)
• Fix for test always returning true (#13911)
• Add error checking for cpp examples. (#13828)
• julia: fix argmax for NDArray (#13871)
• test_ImageRecordIter_seed_augmentation flaky test fix (#12485)
• Julia: fix filename quoting in docstring (#13894)
• Flaky maven binary download (#13974)
• [MXNET-1293] Adding Iterables instead of List to method signature for infer APIs in Java (#13977)
• Sample python bilinear initializer at integral points in y-direction (#12983)
• Fix inconsistent handling for FResourceRequestEx for imperative and symbolic executor (#14007)
• [MXNET-1258] fix unittest for ROIAlign Operator (#13609)
• Fix performance regression in normalize operator (#14055)
• Remove inplace support for ToTensor operator (#14083)
• Addresses comments in runtime feature discovery API (#13964)
• The latest version of leiningen has a dependency problem with codox (#14132)
• Fix quote on LBSGD docs (#13975)
• Fixes spelling (#14168)
• Fix broken amalgamation (#12792)
• Fix nd.pick large array issue (#14082)
• Fix req=null in SliceLikeBackward (#14209)
• onnx broadcast ops fixes (#13604)
• fix update params (#14218)
• MXNet Java bug fixes and experience improvement (#14213)
• reverting broadcasting fixes (#14299)
• fix memory-related issues to enable ASAN tests (#14223)
• FIX: flaky test exponential generator (#14287)
• fix SoftmaxOutput resource bug (#14302)
• Fix shape inference pass (#14153)
• Limit workspace for cudnnGet results (#14326)
• #14199: catch subprocess.CalledProcessError in get_gpus() (#14212)
• Fixes #14181, validate model output shape for ObjectDetector. (#14215)
• Optimizer MXKVStoreUpdater bug fix in serializeState method (#14337)
• Add proper exception message for negative shape in array creation routines (#14362)
• Fix NaN value comparisons in relu, max and min ops (#14262)
• fix engine crash in shutdown phase (#14382)
• Flaky test #14189 (#14190)
• Correct update count with Gluon trainer and update_on_kvstore=False (#14377)
• Fix relative difference scala (#14417)
• fix OOM error during resource allocation (#14444)
• Fix crashes on visualization (#14425)
• begin=end not a valid input (#14403)
• Fix memory leak for size-zero ndarray (#14365)
• Fixes the test_sgld (#14473)
• Revert "Fix memory leak for size-zero ndarray (#14365)" (#14477)
• fix custom operation in fork (#14451)
• Fixes test_operator_gpu.test_multinomial_generator (#14475)
• Fix script retrieval (#14519)
• Memory fixes. Resolves #10867, and resolves #14080 (#14372)
• Chouffe/clojure fix tests (#14531)
• [clojure][image] add draw-bounding-box interop (#14533)
• fix tests (#14565)
• Do not touch GPU 0 during ReleaseAll (#14550)
• [MXNET-1357] Fix the cpp-examples to add exception handling (#14441)
• fix build cpp examples option (#14562)
• Fix flaky test poisson generator & test_negative_binomial_generator (#14571)
• Fixing unintentional variable overloading (#14438)
• fix quantize graph pass (#14605)
• replace std::random_shuffle to std::shuffle (#14523)
• Add exception handling support for waitall (#14397)
• split_and_load can now handle num_ctx > num_data. Issue #13909 (#14607)
• Fix aspect ratio sampling for RandomResizedCrop (#14585)
• [MXNET-400] support string type for kvstore key in cpp-package (#10792)
• Fix warning on macro expansion using defined. (#14598)
• Fix scaladoc scalastyle violations in Infer package (#14671)
• Fix profiler check (#14677)
• Tweak the copy for the cudnn autotuning warning. (#14680)
• Properly handling custom op exception by modify engine (#14693)
• Disable USE_GPERFTOOLS (#14711)
• Reference engine from chunk via weak pointer (#14591)

API Changes

TODO

Depreciations

• Julia: deprecate mx.empty, replace it with UndefInitializer (#13934)
   

Known Issues

How to build MXNet

Please follow the instructions at https://mxnet.incubator.apache.org/install/index.html

List of submodules used by Apache MXNet (Incubating) and when they were updated last