PopTorch 3.4.0 Release Notes PopTorch Geometric 3.4.0 Release Notes

PopTorch 3.3.0 Release Notes PopTorch Geometric 3.3.0 Release Notes

PopTorch 3.2.0 Release Notes PopTorch Geometric 3.2.0 Release Notes

New features

• Upgraded from PyTorch 1.10 to 1.13.
• Added support for variables being sharded across replicas.
• poptorch.set_overlap_for_input and poptorch.set_overlap_for_output can now be applied to tuples, lists, and dicts of tensors.
• PopTorch now catches aten::lstm directly when compiling with dispatch for PopART, allowing set_available_memory to work with it.
• Added support for aten::index_fill_.int_Scalar.
• Added support for dict inputs.
• Added support for torch.count_nonzero.
• Support the tanh approximate for GELU.
• Added support for torch.scatter_reduce operation.

Bug Fixes

• Fixed clamp_max in cases where the max is large.
• Fixed shape inference failing on PopART for argsort, GRU and norm ops.
• Fixed shape inference for strided slices.
• Fixed casting of groupnorm.
• Fixed an issue where the alpha and beta arguments were flipped for torch.addmm.
• Fixed a "not representable" error when using BCEWithLogitsLoss with a dtype of half.
• Fixed intermittent compilation hang caused by tqdm (progress bar).

Other improvements

• Fixed in-place modification of slice regions.
• Documentation typo fixes and clarifications.
• Improved error message when encountering CPU tensors.
• Use the IPU DispatchKey instead of the XLA DispatchKey, which means that error messages will now mention IPU rather XLA.

Known issues

None

Compatibility changes

• Dropped support for Python 3.6 (in order to upgrade to PyTorch 1.13).
• Removed support for torch.jit.trace(). For help on migration issues when using the dispatcher frontend, see the Legacy tracing frontend <https://docs.graphcore.ai/projects/poptorch-user-guide/en/3.0.0/tracing.html>__ section in the 3.0.0 version of the poptorch-user-guide:index.
• Removed support for building on CentOS 7.x.
• Removed the Autocast API (this was only available when using the tracing frontend).

New features

• Support for tensor.new_full
• Improved performance of index_put when the indices are a one dimensional vector.

API changes

• Use the dispatcher instead of torch.jit.trace() by default to capture the graph on supported platforms (see :numref:dispatcher-support). Use poptorch.Options.Jit.traceModel(True) to revert to previous behaviour.

Bug Fixes

• Fixed BatchNorm running statistics, now using the unbiased estimator to update running_var at training time.
• Fixed scalar tensor inputs to the graph not working.
• Fixed expand, when the desired shape contains both added dimensions and -1.

New features

• Improved performance of torch.gather in some cases where the index tensor has come from an expand or expand_as.
• Improved error message when trying to apply bitwise ops to unsupported scalar types.
• Support for upsample bicubic mode.
• Support for zero_infinity in torch.nn.CTCLoss.
• Experimental support for Torch's dispatcher as an alternative to torch.jit.trace().
• Improved performance by compiling built-in custom ops at install time.

API changes

• Data loader: default process creation strategy for worker process changed from "spawn" to "forkserver".

Bug Fixes

• Fixed remaining in-place operations on slices
• Fixed einsum transpose error
• Fixed floating point exception in torch.Tensor.exponential_ and torch.distributions.Exponential
• Improved support for torch.int16 tensors.

New features

• Ignore missing values when reloading an Optimizer state.

• Support saving Optimizer states when compiling offline.

• Support for the following functions:

  • torch.var
  • torch.std
  • torch.var_mean
  • torch.std_mean

• Also save the random number generator's state and the seed when saving a model.

• Support for col2im (used by torch.nn.Fold).

• Improve error message of aten::index, aten::index_put_ when indexing with boolean tensor masks.

• Support for torch.argsort.

• Support for torch.nn.RNN.

• Add support for __repr__ in PoplarExecutor.

• For models annotated with BeginBlock, show the IPU blocks in repr(model).

• Improve implementation of torch.scatter_add

• Support for torch.nn.utils.weight_norm

• Support for torch.randperm

• Support for torch.nn.functional.cosine_similarity and torch.nn.CosineSimilarity

• Support for torch.all, torch.any, torch.Tensor.all and torch.Tensor.any

• Support for torch.Tensor.exponential_ and torch.distributions.Exponential

API changes

• Removed poptorch.AnchorMode, poptorch.Options.anchorMode which were deprecated in favour of poptorch.OutputMode and poptorch.Options.outputMode respectively.

Bug Fixes

• Fix thread safety issue in LogContext.
• Fix torch.clamp with integer tensors
• Fix in-place modification of slices
• Fix torch.index_put_ when operating on slices
• Fix torch.chunk when dim size is indivisible by the specified number of chunks
• Fix cases where tensor.half() was in-place
• Fix tracing with half buffers
• Fix for loops with in-place ops
• Fix torch.flip with negative indices
• Fix masked assign when using tensor indexing syntax
• Fix some cases where use of serializedMatMul was ignored or resulted in errors.

New features

• Support for deepcopy functionality in poptorch.Options class
• Added functionality to add a name scope for each operator present in the module. This function is enabled by default. It can be disabled using poptorch.Options.disableModuleNamescope.
• Support for a greater number of convolution and transpose convolution parameters including those which result in input/kernel/output truncation, either for inference (transpose) or gradient calculation.
• Migrated to PyTorch version 1.10.0
• Support for gradient clipping by norm in poptorch.optim optimizers
• Support saving and restoring internal optimiser state with PopTorch optimisers via optimizer.state_dict() and optimizer.load_state_dict()
• Add removeBlocks function to remove block annotations from a Model / Layer.
• Support for CPU ops using poptorch.CPU.
• Support for im2col.
• Make optimizers work with LR schedulers.
• Switched to gold linker by default.

API changes

• Deprecated poptorch.Options.anchorMode in favour of poptorch.Options.outputMode
• Deprecated poptorch.Options.defaultAnchorMode in favour of poptorch.Options.defaultOutputMode
• Deprecated poptorch.AnchorMode in favour of poptorch.OutputMode

Bug Fixes

• Fixed incorrect gradient when using torch.nn.Embedding with padding_idx

New features

• Support for torch.bitwise_and, torch.bitwise_or, torch.bitwise_xor
• Support for torch.logical_and, torch.logical_or,
• Support K-dimensional NLLLoss, K-dimensional CrossEntropyLoss
• Support for non-default affine parameter flags in normalisation ops
• Support for torch.Tensor.T
• Support for torch.bool in torch.zeros, torch.zeros_like, torch.ones, torch.ones_like
• Support for torch.scatter and its in-place variant
• Support for in-place modification to buffers on IPU
• Support for taking slices of scalars
• Support version of bilinear upsampling specifying intended output size instead of scale factors
• Add support for overlapping host IO on inputs via poptorch.set_overlap_for_input().
• Add option for setting number of IO tiles via numIOTiles in poptorch.Options (required for useIOTilesToLoad() and - poptorch.set_overlap_for_input().)
• Improve PopTorch’s parity with PyTorch’s Softplus
• Improve implementation of torch.SiLU by using Poplar’s Swish operator
• Additional support for operation overloads
• Fix issue where PopTorch recalculated upsampling scales in fp16
• Fix issue where the last use of poptorch.set_available_memory would be pruned
• Add documentation on available memory proportion to incorporate embeddings and indexing operations
• Add documentation on how users can generate debug information
• Support replicated tensor sharding when running on multiple processes
• Allow selection for a non-constant x input.
• Support for enableConvDithering convolution option

API changes

• Default mean reduction strategies have changed from the deprecated PostAndLoss strategy to Post or Running based on optimiser accumulation type
• Mean reduction strategy can now be set via poptorch.Options.Training.setMeanAccumulationAndReplicationReductionStrategy.
• Add warning that IPU-specific optimiser states cannot be read from the host, when calling get_state() on poptorch.optim optimisers

New features

• Migrated to PyTorch version 1.9.0
• Support for torch.roll
• Support for torch.clone
• Add modelName session option that can be passed to PopART
• Support List inputs to a model
• Tuples/Lists of constants can now be returned by a model
• Add enableProfiling convenience method in poptorch.Options to enable profile report generation
• Fix bug with torch.Tensor.repeat when applied to an input during training
• Fix bug with aten::to when applied to a constant used as an input to another node
• Improved error message when encountering untraceable types during compilation
• Support for torch.gather. Please note: this operator is known to cause long compilation times. Consider using a onehot-based solution instead or torch.index_select if appropriate.
• Using a convolution layer op with the value of padding greater than or equal to kernel_size` is now supported.
• Support for Poplar recoverable and unrecoverable errors.
• Support for torch.flip.
• Support for torch.Tensor.new_ones and torch.Tensor.new_zeros

API changes

• Removed accumulationReductionType which was deprecated in 2.1 in favour of accumulationAndReplicationReductionType in poptorch.Options.Training
• Removed runningVarianceAlwaysFloat which was deprecated in 2.1 and replaced by runningStatisticsAlwaysFloat in poptorch.Options.Precision.