Targeting the IPU from TensorFlow 1

• 1. Introduction
  • 1.1. Document overview
• 2. Tutorial
  • 2.1. Preliminary graphs
  • 2.2. A basic graph
    • 2.2.1. Selecting hardware to run on
    • 2.2.2. Running on the IPU Model simulator
  • 2.3. Compiling the graph for the IPU
  • 2.4. Sharding a graph
  • 2.5. Adding variables
    • 2.5.1. Troubleshooting
    • 2.5.2. Note on the global_step counter
• 3. Targeting the Poplar XLA device
  • 3.1. Supported types
  • 3.2. Device selection
  • 3.3. Configuring system options
    • 3.3.1. TF_POPLAR_FLAGS environment variable
  • 3.4. Supported operations
  • 3.5. Unsupported operations
  • 3.6. Error Handling
    • 3.6.1. Construction and compilation errors
    • 3.6.2. Runtime errors
• 4. Compiling and pre-compiling executables
  • 4.1. Caching of compiled executables
  • 4.2. Pre-compiling executables
    • 4.2.1. Unsupported Operations
• 5. Training a model
  • 5.1. Training loops, data sets and feed queues
  • 5.2. Accessing outfeed queue results during execution
  • 5.3. Replicated graphs
    • 5.3.1. Selecting the number of replicas
    • 5.3.2. Performing parameter updates
  • 5.4. Pipelined training
    • 5.4.1. Grouped scheduling
    • 5.4.2. Interleaved scheduling
    • 5.4.3. Sequential scheduling
    • 5.4.4. Pipeline stage inputs and outputs
    • 5.4.5. Applying an optimiser to the graph
    • 5.4.6. Device mapping
    • 5.4.7. Concurrent pipeline stages
  • 5.5. Gradient accumulation
    • 5.5.1. Optimizers
    • 5.5.2. Pipelining
    • 5.5.3. Accumulation data type
  • 5.6. Optimizer state offloading
  • 5.7. Dataset benchmarking
    • 5.7.1. Accessing the JSON data
  • 5.8. Half and mixed precision training
• 6. Efficient IPU I/O
  • 6.1. Prefetch elements
  • 6.2. I/O Tiles
• 7. Example using IPUEstimator
• 8. Example using IPUPipelineEstimator
• 9. Distributed training
  • 9.1. PopDistStrategy examples
• 10. Half-precision floating point and stochastic rounding
  • 10.1. Controlling the half-precision floating-point unit
  • 10.2. Resetting the global random number seed
  • 10.3. Debugging numerical issues
• 11. IPU-optimised operations
  • 11.1. Image operations
  • 11.2. Matmul serialisation
  • 11.3. Dropout
  • 11.4. Embedding lookup
  • 11.5. Group normalisation
  • 11.6. Instance normalisation
  • 11.7. Layer normalisation
  • 11.8. GeLU activation
  • 11.9. Sequence slice
  • 11.10. Histogram
• 12. IPU Outlined Functions
  • 12.1. Usage
  • 12.2. Examples
    • 12.2.1. Models with common structures
    • 12.2.2. Serializing large operations
• 13. Writing custom operations
  • 13.1. Custom operation on the IPU
    • 13.1.1. Building the Poplar graph
    • 13.1.2. Gradient builders
    • 13.1.3. Metadata
    • 13.1.4. Compiling the IPU code
      • API level
      • PopLibs library code
      • Compiling the library file
    • 13.1.5. Using the custom op in TensorFlow
    • 13.1.6. Tensor allocation
    • 13.1.7. Examples
      • In-place operations
      • Operation attributes
      • Custom codelet
  • 13.2. Custom host CPU operations
    • 13.2.1. Gradient callback
• 14. IPU host embeddings
  • 14.1. Usage
  • 14.2. Example
  • 14.3. Experimental functionality: IPU embeddings in remote buffers
    • 14.3.1. Partitioning strategies
      • Token strategy
      • Encoding strategy
      • Choosing a strategy for your application
• 15. IPU embedded application runtime
  • 15.1. Usage
  • 15.2. Pipelining and I/O tiles
    • 15.2.1. Parallel requests
    • 15.2.2. Timeout
    • 15.2.3. Engine restarts
  • 15.3. Example
  • 15.4. Error Handling
    • 15.4.1. Runtime errors
• 16. Exporting precompiled models for TensorFlow Serving
  • 16.1. Exporting non-pipelined models defined inside a function
    • 16.1.1. Example of exporting non-pipelined model defined inside a function
    • 16.1.2. Example of exporting non-pipelined model defined inside a function with additional preprocessing and postprocessing steps
  • 16.2. Exporting pipelined models defined as a list of functions
    • 16.2.1. Pipeline example
    • 16.2.2. Pipeline example with preprocessing and postprocessing steps
  • 16.3. Running the model in TensorFlow Serving
• 17. Retrieving information about compilation and execution
  • 17.1. TensorFlow options for reporting
  • 17.2. XLA graph file naming
• 18. IPU TensorFlow Addons
  • 18.1. Introduction
  • 18.2. IPU SavedModel CLI
    • 18.2.1. Run subcommand
    • 18.2.2. Convert subcommand
    • 18.2.3. Pipeline configuration
    • 18.2.4. Pipeline development
    • 18.2.5. Pipeline solution file
    • 18.2.6. Example configuration file
• 19. TensorFlow API changes
  • 19.1. Release 3.0
    • 19.1.1. Non-breaking changes
      • Deprecated modules
  • 19.2. Release 2.6
    • 19.2.1. Breaking changes
      • Removal of deprecated APIs
  • 19.3. Release 2.5
    • 19.3.1. Breaking changes
      • Removal of deprecated APIs
      • Other
    • 19.3.2. Non-breaking changes
      • Deprecated layers
      • RNN available_memory_proportion_fwd/available_memory_proportion_bwd deprecated
  • 19.4. Release 2.4
    • 19.4.1. Breaking changes
      • Summary ops
      • Removal of deprecated members
    • 19.4.2. Non-breaking changes
  • 19.5. Release 2.3
    • 19.5.1. Breaking changes
      • Custom user op metadata interface updates
      • The verified transfers feature has been removed
    • 19.5.2. Non-breaking changes
  • 19.6. Release 2.2
    • 19.6.1. Breaking changes
      • C++ Poplar TensorFlow libraries are private by default
      • Reports removed from ipu events
    • 19.6.2. Non-breaking changes
      • IPULoggingTensorHook replication_factor deprecated
      • IPUInfeedQueue/IPUOutfeedQueue/IPULoggingTensorHook feed_name deprecated
      • Change of output location for profiling information
      • IPU Keras Layers deprecation in TensorFlow 1.15
      • Warning when epsilon value is too low
  • 19.7. Release 2.1
    • 19.7.1. Breaking changes
      • IPUPipelineEstimator change
      • Autosharding removed
      • Old IPU option configuration API changes
      • IPU Keras changes [TensorFlow 2]
    • 19.7.2. Non-breaking changes
      • Recompute suggestions deprecated
      • IPUInfeedQueue/IPUOutfeedQueue replication_factor deprecated
      • IPUInfeedQueue data_to_prefetch deprecated
      • IPUOutfeedQueue data_to_prefetch deprecated
      • CTC loss ops deprecated
      • New configuration API
      • Support for grouped collectives
      • Environment variable changes
  • 19.8. Release 2.0
    • 19.8.1. Breaking changes
    • 19.8.2. Non-breaking changes
      • IPUPipelineEstimator change
      • Autosharding deprecated
      • IPU config change
      • IPU Keras changes [TensorFlow 2]
• 20. TensorFlow Python API
  • 20.1. Operations and utilities related to the Graphcore IPU
  • 20.2. Compiler interface
  • 20.3. Scoping contexts
  • 20.4. Infeed queue
  • 20.5. Outfeed queue
  • 20.6. General utilities
  • 20.7. Configuration utilities
  • 20.8. Looping utilities
  • 20.9. Distributed training
  • 20.10. Horovod
  • 20.11. Serving utilities
  • 20.12. Datasets
    • 20.12.1. Dataset benchmarking
    • 20.12.2. Dataset wrappers
  • 20.13. Estimators
    • 20.13.1. IPUEstimator
    • 20.13.2. IPUPipelineEstimator
    • 20.13.3. Run configs
    • 20.13.4. Session run hooks
  • 20.14. Keras layers
    • 20.14.1. Keras layer specializations for the Graphcore IPU
  • 20.15. Operators
    • 20.15.1. Control flow operations.
    • 20.15.2. Custom operations
    • 20.15.3. Functional operators
    • 20.15.4. Image operations
    • 20.15.5. Graphcore utility operations
    • 20.15.6. IPU specific maths operations
    • 20.15.7. Pipelining operators
    • 20.15.8. Popnn primitive neural network operators
    • 20.15.9. Popnn normalization operators
    • 20.15.10. Popops all to all and all gather operators
    • 20.15.11. Popops cross replica operators
    • 20.15.12. Popops embedding operators
    • 20.15.13. Popops reduce scatter operator
    • 20.15.14. Popops within replica operators
    • 20.15.15. Poprand operators
    • 20.15.16. Utility operations to be used in replicated mode
    • 20.15.17. Slicing operators
    • 20.15.18. Statistics operators
    • 20.15.19. Embedded application runtime
  • 20.16. Optimisers
    • 20.16.1. Helper classes and methods for gradient accumulation.
    • 20.16.2. Optimizer classes for the Graphcore IPU
  • 20.17. Sharding
    • 20.17.1. Utility functions for sharding graphs
• 21. TensorFlow operators supported by the IPU
• 22. IPU TensorFlow Addons API changes
  • 22.1. Release 3.0
    • 22.1.1. Breaking changes
  • 22.2. Release 2.5
    • 22.2.1. Non-breaking changes
      • RNN available_memory_proportion_fwd/available_memory_proportion_bwd deprecated
  • 22.3. Release 2.4
• 23. IPU TensorFlow Addons Python API
  • 23.1. TensorFlow layers
    • 23.1.1. TensorFlow layers made for IPU TensorFlow
  • 23.2. TensorFlow optimizers
    • 23.2.1. Optimizers made for IPU TensorFlow
• 24. Resources
  • 24.1. Graphcore
  • 24.2. TensorFlow
  • 24.3. Other
• 25. Trademarks & copyright