diff options
Diffstat (limited to 'tensorflow/docs_src')
| -rw-r--r-- | tensorflow/docs_src/community/lists.md | 2 | ||||
| -rw-r--r-- | tensorflow/docs_src/guide/eager.md | 3 | ||||
| -rw-r--r-- | tensorflow/docs_src/guide/version_compat.md | 1 | ||||
| -rw-r--r-- | tensorflow/docs_src/install/install_go.md | 2 | ||||
| -rw-r--r-- | tensorflow/docs_src/install/install_raspbian.md | 6 | ||||
| -rw-r--r-- | tensorflow/docs_src/install/install_sources.md | 5 | ||||
| -rw-r--r-- | tensorflow/docs_src/performance/xla/operation_semantics.md | 97 |
7 files changed, 98 insertions, 18 deletions
diff --git a/tensorflow/docs_src/community/lists.md b/tensorflow/docs_src/community/lists.md index 7450ab36c4..bc2f573c29 100644 --- a/tensorflow/docs_src/community/lists.md +++ b/tensorflow/docs_src/community/lists.md @@ -32,6 +32,8 @@ These projects inside the TensorFlow GitHub organization have lists dedicated to and peer support for TensorFlow.js. * [tflite](https://groups.google.com/a/tensorflow.org/d/forum/tflite) - Discussion and peer support for TensorFlow Lite. +* [tfprobability](https://groups.google.com/a/tensorflow.org/d/forum/tfprobability) - Discussion and + peer support for TensorFlow Probability. * [tpu-users](https://groups.google.com/a/tensorflow.org/d/forum/tpu-users) - Community discussion and support for TPU users. diff --git a/tensorflow/docs_src/guide/eager.md b/tensorflow/docs_src/guide/eager.md index 24f6e4ee95..017fdaf81e 100644 --- a/tensorflow/docs_src/guide/eager.md +++ b/tensorflow/docs_src/guide/eager.md @@ -193,8 +193,7 @@ class MNISTModel(tf.keras.Model): def call(self, input): """Run the model.""" result = self.dense1(input) - result = self.dense2(result) - result = self.dense2(result) # reuse variables from dense2 layer + result = self.dense2(result) # reuse variables from dense1 layer return result model = MNISTModel() diff --git a/tensorflow/docs_src/guide/version_compat.md b/tensorflow/docs_src/guide/version_compat.md index 0e472c3381..29ac066e6f 100644 --- a/tensorflow/docs_src/guide/version_compat.md +++ b/tensorflow/docs_src/guide/version_compat.md @@ -79,6 +79,7 @@ backward incompatible ways between minor releases. These include: [`tensorflow/cc`](https://github.com/tensorflow/tensorflow/tree/master/tensorflow/cc)). - [Java](../api_docs/java/reference/org/tensorflow/package-summary), - [Go](https://godoc.org/github.com/tensorflow/tensorflow/tensorflow/go) + - [JavaScript](https://js.tensorflow.org) * **Details of composite ops:** Many public functions in Python expand to several primitive ops in the graph, and these details will be part of any diff --git a/tensorflow/docs_src/install/install_go.md b/tensorflow/docs_src/install/install_go.md index 961386b15c..f0f8436777 100644 --- a/tensorflow/docs_src/install/install_go.md +++ b/tensorflow/docs_src/install/install_go.md @@ -6,7 +6,7 @@ a Go application. This guide explains how to install and set up the [TensorFlow Go package](https://godoc.org/github.com/tensorflow/tensorflow/tensorflow/go). Warning: The TensorFlow Go API is *not* covered by the TensorFlow -[API stability guarantees](../guide/version_semantics.md). +[API stability guarantees](../guide/version_compat.md). ## Supported Platforms diff --git a/tensorflow/docs_src/install/install_raspbian.md b/tensorflow/docs_src/install/install_raspbian.md index 58a5285c78..cf6b6b4f79 100644 --- a/tensorflow/docs_src/install/install_raspbian.md +++ b/tensorflow/docs_src/install/install_raspbian.md @@ -60,7 +60,7 @@ If it gives the error "Command not found", then the package has not been installed yet. To install if for the first time, run: <pre>$ sudo apt-get install python3-pip # for Python 3.n -sudo apt-get install python-pip # for Python 2.7</pre> +$ sudo apt-get install python-pip # for Python 2.7</pre> You can find more help on installing and upgrading pip in [the Raspberry Pi documentation](https://www.raspberrypi.org/documentation/linux/software/python.md). @@ -78,8 +78,8 @@ your system, run the following command: Assuming the prerequisite software is installed on your Pi, install TensorFlow by invoking **one** of the following commands: - <pre> $ <b>pip3 install tensorflow</b> # Python 3.n - $ <b>pip install tensorflow</b> # Python 2.7</pre> +<pre>$ <b>pip3 install tensorflow</b> # Python 3.n +$ <b>pip install tensorflow</b> # Python 2.7</pre> This can take some time on certain platforms like the Pi Zero, where some Python packages like scipy that TensorFlow depends on need to be compiled before the diff --git a/tensorflow/docs_src/install/install_sources.md b/tensorflow/docs_src/install/install_sources.md index 1d108d8bd5..dfd9fbce4b 100644 --- a/tensorflow/docs_src/install/install_sources.md +++ b/tensorflow/docs_src/install/install_sources.md @@ -180,7 +180,10 @@ If you follow these instructions, you will not need to disable SIP. After installing pip, invoke the following commands: -<pre> $ <b>sudo pip install six numpy wheel mock</b> </pre> +<pre> $ <b>sudo pip install six numpy wheel mock h5py</b> + $ <b>sudo pip install keras_applications==1.0.4 --no-deps</b> + $ <b>sudo pip install keras_preprocessing==1.0.2 --no-deps</b> +</pre> Note: These are just the minimum requirements to _build_ tensorflow. Installing the pip package will download additional packages required to _run_ it. If you diff --git a/tensorflow/docs_src/performance/xla/operation_semantics.md b/tensorflow/docs_src/performance/xla/operation_semantics.md index 165f6f5914..fad9fd57f1 100644 --- a/tensorflow/docs_src/performance/xla/operation_semantics.md +++ b/tensorflow/docs_src/performance/xla/operation_semantics.md @@ -13,6 +13,79 @@ arbitrary-dimensional array. For convenience, special cases have more specific and familiar names; for example a *vector* is a 1-dimensional array and a *matrix* is a 2-dimensional array. +## AllToAll + +See also +[`XlaBuilder::AllToAll`](https://www.tensorflow.org/code/tensorflow/compiler/xla/client/xla_builder.h). + +Alltoall is a collective operation that sends data from all cores to all cores. +It has two phases: + +1. the scatter phase. On each core, the operand is split into `split_count` + number of blocks along the `split_dimensions`, and the blocks are scatterd + to all cores, e.g., the ith block is send to the ith core. +2. the gather phase. Each core concatenates the received blocks along the + `concat_dimension`. + +The participating cores can be configured by: + +- `replica_groups`: each ReplicaGroup contains a list of replica id. If empty, + all replicas belong to one group in the order of 0 - (n-1). Alltoall will be + applied within subgroups in the specified order. For example, replica + groups = {{1,2,3},{4,5,0}} means, an Alltoall will be applied within replica + 1, 2, 3, and in the gather phase, the received blocks will be concatenated + in the order of 1, 2, 3; another Alltoall will be applied within replica 4, + 5, 0, and the concatenation order is 4, 5, 0. + +Prerequisites: + +- The dimension size of the operand on the split_dimension is divisible by + split_count. +- The operand's shape is not tuple. + +<b> `AllToAll(operand, split_dimension, concat_dimension, split_count, +replica_groups)` </b> + + +| Arguments | Type | Semantics | +| ------------------ | --------------------- | ------------------------------- | +| `operand` | `XlaOp` | n dimensional input array | +| `split_dimension` | `int64` | A value in the interval `[0, | +: : : n)` that names the dimension : +: : : along which the operand is : +: : : split : +| `concat_dimension` | `int64` | a value in the interval `[0, | +: : : n)` that names the dimension : +: : : along which the split blocks : +: : : are concatenated : +| `split_count` | `int64` | the number of cores that | +: : : participate this operation. If : +: : : `replica_groups` is empty, this : +: : : should be the number of : +: : : replicas; otherwise, this : +: : : should be equal to the number : +: : : of replicas in each group. : +| `replica_groups` | `ReplicaGroup` vector | each group contains a list of | +: : : replica id. : + +Below shows an example of Alltoall. + +``` +XlaBuilder b("alltoall"); +auto x = Parameter(&b, 0, ShapeUtil::MakeShape(F32, {4, 16}), "x"); +AllToAll(x, /*split_dimension=*/1, /*concat_dimension=*/0, /*split_count=*/4); +``` + +<div style="width:95%; margin:auto; margin-bottom:10px; margin-top:20px;"> + <img style="width:100%" src="../../images/xla/ops_alltoall.png"> +</div> + +In this example, there are 4 cores participating the Alltoall. On each core, the +operand is split into 4 parts along dimension 0, so each part has shape +f32[4,4]. The 4 parts are scattered to all cores. Then each core concatenates +the received parts along dimension 1, in the order or core 0-4. So the output on +each core has shape f32[16,4]. + ## BatchNormGrad See also @@ -1804,19 +1877,19 @@ See also [`XlaBuilder::RngNormal`](https://www.tensorflow.org/code/tensorflow/compiler/xla/client/xla_builder.h). Constructs an output of a given shape with random numbers generated following -the $$N(\mu, \sigma)$$ normal distribution. The parameters `mu` and `sigma`, and -output shape have to have elemental type F32. The parameters furthermore have to -be scalar valued. +the $$N(\mu, \sigma)$$ normal distribution. The parameters $$\mu$$ and +$$\sigma$$, and output shape have to have a floating point elemental type. The +parameters furthermore have to be scalar valued. -<b>`RngNormal(mean, sigma, shape)`</b> +<b>`RngNormal(mu, sigma, shape)`</b> | Arguments | Type | Semantics | | --------- | ------- | --------------------------------------------------- | -| `mu` | `XlaOp` | Scalar of type F32 specifying mean of generated | -: : : numbers : -| `sigma` | `XlaOp` | Scalar of type F32 specifying standard deviation of | +| `mu` | `XlaOp` | Scalar of type T specifying mean of generated | +: : : numbers : +| `sigma` | `XlaOp` | Scalar of type T specifying standard deviation of | : : : generated numbers : -| `shape` | `Shape` | Output shape of type F32 | +| `shape` | `Shape` | Output shape of type T | ## RngUniform @@ -1825,9 +1898,11 @@ See also Constructs an output of a given shape with random numbers generated following the uniform distribution over the interval $$[a,b)$$. The parameters and output -shape may be either F32, S32 or U32, but the types have to be consistent. -Furthermore, the parameters need to be scalar valued. If $$b <= a$$ the result -is implementation-defined. +element type have to be a boolean type, an integral type or a floating point +types, and the types have to be consistent. The CPU and GPU backends currently +only support F64, F32, F16, BF16, S64, U64, S32 and U32. Furthermore, the +parameters need to be scalar valued. If $$b <= a$$ the result is +implementation-defined. <b>`RngUniform(a, b, shape)`</b> |