Commit message (Collapse) | Author | Age | |
---|---|---|---|
* | TFLite: Rename ResetVariableTensorsToZero -> ResetVariableTensors | Yu-Cheng Ling | 2018-09-27 |
| | | | | PiperOrigin-RevId: 214820383 | ||
* | Add an experimental API to allow half precision for FP32 calculation. | A. Unique TensorFlower | 2018-09-11 |
| | | | | PiperOrigin-RevId: 212535448 | ||
* | Modularize TF Lite interface definitions and reorganize file structure | Pete Warden | 2018-09-07 |
| | | | | PiperOrigin-RevId: 212064501 | ||
* | Preallocate the node structure std::vector | Andrew Selle | 2018-08-22 |
| | | | | PiperOrigin-RevId: 209830234 | ||
* | Provide TfLiteContext arg to all TfLiteDelegate callbacks | Jared Duke | 2018-08-14 |
| | | | | | | | Note: This change may break clients who have custom TfLiteDelegate implementations; this API has been and remains experimental and subject to such changes. PiperOrigin-RevId: 208683190 | ||
* | Support TFLite framework compilation on Windows | Jared Duke | 2018-08-01 |
| | | | | PiperOrigin-RevId: 207020196 | ||
* | An ErrorReporter to be used in tests. | A. Unique TensorFlower | 2018-07-25 |
| | | | | PiperOrigin-RevId: 206012444 | ||
* | Small changes to interpreter.{h,cc}: refactoring plus improved error message. | A. Unique TensorFlower | 2018-07-25 |
| | | | | PiperOrigin-RevId: 205992521 | ||
* | More info on op prepare/eval failures | A. Unique TensorFlower | 2018-07-17 |
| | | | | PiperOrigin-RevId: 204932609 | ||
* | Make an option that allows making the memory plan not reuse space. | Andrew Selle | 2018-07-13 |
| | | | | | | | | This is called 'allow_intermediates'. In the future we should have a way for users to enable this functionality from the interpreter API. Also, preserve_inputs is now better commented. PiperOrigin-RevId: 204496777 | ||
* | Reset variable tensors to zero after allocating tensors. | Yu-Cheng Ling | 2018-07-10 |
| | | | | PiperOrigin-RevId: 204022562 | ||
* | Make tensorflow lite interpreter have more helpful errors. | Andrew Selle | 2018-07-10 |
| | | | | PiperOrigin-RevId: 203961654 | ||
* | Remove framework's dependency on eigen and gemmlowp. | A. Unique TensorFlower | 2018-07-03 |
| | | | | PiperOrigin-RevId: 203172717 | ||
* | Avoid redundant TFLite tensor reallocations | Jared Duke | 2018-07-02 |
| | | | | PiperOrigin-RevId: 202988873 | ||
* | Do not overwrite inputs. | A. Unique TensorFlower | 2018-06-29 |
| | | | | PiperOrigin-RevId: 202724720 | ||
* | Add support for Makefile build for micro-controller platforms | A. Unique TensorFlower | 2018-06-29 |
| | | | | PiperOrigin-RevId: 202716942 | ||
* | Add complex64 support to tf.lite runtime. | RJ Ryan | 2018-06-27 |
| | | | | PiperOrigin-RevId: 202403235 | ||
* | Refresh allocations in the presence of dynamic tensors | A. Unique TensorFlower | 2018-06-19 |
| | | | | PiperOrigin-RevId: 201193941 | ||
* | Variable Tensor API for TF Lite. | Yu-Cheng Ling | 2018-06-13 |
| | | | | PiperOrigin-RevId: 200457602 | ||
* | support int16-quantized data in TFLite interpreter. | A. Unique TensorFlower | 2018-06-13 |
| | | | | PiperOrigin-RevId: 200442886 | ||
* | Small fix to prevent a crash if the delegate has not implemented ↵ | A. Unique TensorFlower | 2018-04-30 |
| | | | | | | FreeBufferHandle. PiperOrigin-RevId: 194866595 | ||
* | Convert int -> size_t so that implicit conversion doesn't lose integer ↵ | A. Unique TensorFlower | 2018-04-23 |
| | | | | | | precision. PiperOrigin-RevId: 193955175 | ||
* | Add a simple Profiler and instrument operator invocations in Interpreter. | Shashi Shekhar | 2018-04-16 |
| | | | | PiperOrigin-RevId: 193133955 | ||
* | Internal change. | A. Unique TensorFlower | 2018-04-16 |
| | | | | PiperOrigin-RevId: 193112205 | ||
* | TFLite: Copy output data from BufferHandle to CPU memory by default. | Yu-Cheng Ling | 2018-04-13 |
| | | | | PiperOrigin-RevId: 192846824 | ||
* | Add boolean type to tflite in favor of comparison implementations. | A. Unique TensorFlower | 2018-04-12 |
| | | | | PiperOrigin-RevId: 192711203 | ||
* | Revert to initializing number of threads when SetNumThreads is called. ↵ | A. Unique TensorFlower | 2018-03-26 |
| | | | | | | | | Requiring it to happen before OpInit() is way too confusing for users. PiperOrigin-RevId: 190499644 | ||
* | TFLite Delegate: Add an `allow_dynamic_tensors` parameter. | Yu-Cheng Ling | 2018-03-19 |
| | | | | PiperOrigin-RevId: 189641833 | ||
* | TFLite Delegate: Expose input / output tensor indicies in `Init` | Yu-Cheng Ling | 2018-03-12 |
| | | | | PiperOrigin-RevId: 188784614 | ||
* | [TFLite] Don't require a std::vector for Interpreter::SetTensorParameters*. | Eugene Brevdo | 2018-03-12 |
| | | | | PiperOrigin-RevId: 188770522 | ||
* | Make default number of threads trigger the default behavior for both eigen | A. Unique TensorFlower | 2018-03-12 |
| | | | | | | | and gemmlowp. In gemmlowp the default is '1', while in eigen it is 'number of processors'. PiperOrigin-RevId: 188742087 | ||
* | Removed duplicate statement. | A. Unique TensorFlower | 2018-03-11 |
| | | | | PiperOrigin-RevId: 188663018 | ||
* | Make SetNumThreads apply to the eigen threads. (This creates a dependency on ↵ | A. Unique TensorFlower | 2018-03-09 |
| | | | | | | eigen!) PiperOrigin-RevId: 188504172 | ||
* | Return kTfLiteError if calling delegate-specific functions from non-delegate ↵ | Yu-Cheng Ling | 2018-03-08 |
| | | | | | | code. PiperOrigin-RevId: 188407931 | ||
* | [TF Lite] Provide a fast path for Interpreter::SetTensorParametersReadOnly. | Eugene Brevdo | 2018-03-08 |
| | | | | | | | | | | The fast path kicks in if shape matches tensor.dims and type matches tensor.type. In this case, the interpreter is not invalidated and AllocateTensors need not be called after a call to SetTensorParametersReadOnly. PiperOrigin-RevId: 188380596 | ||
* | TFLite: Delegate Buffer Handle interface (take 2) | Yu-Cheng Ling | 2018-03-08 |
| | | | | PiperOrigin-RevId: 188366045 | ||
* | Automated g4 rollback of changelist 188263046 | Gunhan Gulsoy | 2018-03-08 |
| | | | | PiperOrigin-RevId: 188293315 | ||
* | TFLite: Delegate Buffer Handle interface | Yu-Cheng Ling | 2018-03-07 |
| | | | | PiperOrigin-RevId: 188263046 | ||
* | TFLite: Ensures pointers to tensors won't be invalidated unless 16+ tensors ↵ | Yu-Cheng Ling | 2018-02-26 |
| | | | | | | are added. PiperOrigin-RevId: 187052100 | ||
* | TFLite: Define a DELEGATE op type. | Yu-Cheng Ling | 2018-02-20 |
| | | | | PiperOrigin-RevId: 186405366 | ||
* | Fill the new `custom_initial_data(_size)?` fields in TfLiteNode. | Yu-Cheng Ling | 2018-02-20 |
| | | | | PiperOrigin-RevId: 186389819 | ||
* | Fix bug in populating the execution plan sent to the delegate. | Andrew Selle | 2018-02-13 |
| | | | | | | | | - memcpy was missing the array size. - modified the unit test to verify that the execution plan is trivial on first delegate invocation. PiperOrigin-RevId: 185569606 | ||
* | Add delegate API to tflite. | Andrew Selle | 2018-02-09 |
| | | | | | | | | | - Context gets GetNodes, num_nodes and PartitionNodesIntoSubgraphs. - TfLiteDelegate provides one function that need be implemented - Delegates choose nodes and those nodes are all compacted into a new macro kernel. PiperOrigin-RevId: 185204338 | ||
* | make calling NNAPI work again | A. Unique TensorFlower | 2018-02-07 |
| | | | | | | (this a copy of #16256 which apparently got lost in the one of the merges) PiperOrigin-RevId: 184866202 | ||
* | Proper reallocation of dynamic tensors. | A. Unique TensorFlower | 2018-02-05 |
| | | | | PiperOrigin-RevId: 184550199 | ||
* | Allow reordering of execution order of nodes with indirect execution_plan. | Andrew Selle | 2018-02-01 |
| | | | | | | | | Now whenever we want to operate in dependency order we use execution_plan. It begins as identity map (0, ..., nodes_size()) but can be changed in the future. This is the basis for more pluggable delegation. PiperOrigin-RevId: 184216885 | ||
* | Clean up the allocation logic in the interpreter. | A. Unique TensorFlower | 2018-01-13 |
| | | | | PiperOrigin-RevId: 181865795 | ||
* | Internal Change. | Andrew Selle | 2017-11-10 |
PiperOrigin-RevId: 175307445 |