TensorFlow v0.12.0 RC0 发布了。

主要新特性和改进：

• TensorFlow now builds and runs on Microsoft Windows (tested on Windows 10, Windows 7, and Windows Server 2016). Supported languages include Python (via a pip package) and C++. CUDA 8.0 and cuDNN 5.1 are supported for GPU acceleration. Known limitations include: It is not currently possible to load a custom op library. The GCS and HDFS file systems are not currently supported. The following ops are not currently implemented: DepthwiseConv2dNative, DepthwiseConv2dNativeBackpropFilter, DepthwiseConv2dNativeBackpropInput, Dequantize, Digamma, Erf, Erfc, Igamma, Igammac, Lgamma, Polygamma, QuantizeAndDequantize, QuantizedAvgPool, QuantizedBatchNomWithGlobalNormalization, QuantizedBiasAdd, QuantizedConcat, QuantizedConv2D, QuantizedMatmul, QuantizedMaxPool, QuantizeDownAndShrinkRange, QuantizedRelu, QuantizedRelu6, QuantizedReshape, QuantizeV2, RequantizationRange, and Requantize.

• Go: Experimental API in Go to create and execute graphs (https://godoc.org/github.com/tensorflow/tensorflow/tensorflow/go)

• New checkpoint format becomes the default in tf.train.Saver. Old V1 checkpoints continue to be readable; controlled by the write_version argument, tf.train.Saver now by default writes out in the new V2 format. It significantly reduces the peak memory required and latency incurred during restore.

• Added a new library for library of matrix-free (iterative) solvers for linear equations, linear least-squares, eigenvalues and singular values in tensorflow/contrib/solvers. Initial version has lanczos bidiagonalization, conjugate gradients and CGLS.

• Added gradients for matrix_solve_ls and self_adjoint_eig.

• Large cleanup to add second order gradient for ops with C++ gradients and improve existing gradients such that most ops can now be differentiated multiple times.

• Added a solver for ordinary differential equations, tf.contrib.integrate.odeint.

• New contrib module for tensors with named axes, tf.contrib.labeled_tensor.

• Visualization of embeddings in TensorBoard.

API的重大改变：

• BusAdjacency enum replaced with a protocol buffer DeviceLocality. PCI bus indexing now starts from 1 instead of 0, and bus_id==0 is used where previously BUS_ANY was used.

• Env::FileExists and FileSystem::FileExists now return a tensorflow::Status intead of a bool. Any callers to this function can be converted to a bool by adding .ok() to the call.

• C API: Type TF_SessionWithGraph has been renamed to TF_Session, indicating its preferred use in language bindings for TensorFlow. What was previously TF_Session has been renamed toTF_DeprecatedSession.

• C API: Renamed TF_Port to TF_Output.

• C API: The caller retains ownership of TF_Tensor objects provided to TF_Run, TF_SessionRun,TF_SetAttrTensor etc.

• Renamed tf.image.per_image_whitening() to tf.image.per_image_standardization()

• Move Summary protobuf constructors to tf.summary submodule.

• Deprecate histogram_summary, audio_summary, scalar_summary, image_summary, merge_summary, and merge_all_summaries.

• Combined batch_* and regular version of linear algebra and FFT ops. The regular op now handles batches as well. All batch_* Python interfaces were removed.

• tf.all_variables, tf.VARIABLES and tf.initialize_all_variables renamed totf.global_variables, tf.GLOBAL_VARIABLES and tf.global_variable_initializers respectively.

重大修复和其他更新

• Use threadsafe version of lgamma function.

• Fix tf.sqrt handling of negative arguments.

• Fixed bug causing incorrect number of threads to be used for multi-threaded benchmarks.

• Performance optimizations for batch_matmul on multi-core CPUs.

• Improve trace, matrix_set_diag, matrix_diag_part and their gradients to work for rectangular matrices.

• Support for SVD of complex valued matrices

更多详情请点击更新日志。

下载地址：

• Source code (zip)

• Source code (tar.gz)