Merge pull request #5 from tensorflow/master

Stay up to date

CONTRIBUTING.md

@@ -200,6 +200,12 @@ There are two ways to run TensorFlow unit tests.
     bazel test ${flags} //tensorflow/python/...
     ```
 
+    For a single component e.g. softmax op:
+
+    ```bash
+    bazel test ${flags} tensorflow/python/kernel_tests:softmax_op_test
+    ```
+
 2.  Using [Docker](https://www.docker.com) and TensorFlow's CI scripts.
 
     ```bash

README.md

@@ -165,7 +165,7 @@ Container Type                                                    | Status | Art
 *   [Learn ML with TensorFlow](https://www.tensorflow.org/resources/learn-ml)
 *   [TensorFlow Twitter](https://twitter.com/tensorflow)
 *   [TensorFlow YouTube](https://www.youtube.com/channel/UC0rqucBdTuFTjJiefW5t-IQ)
-*   [TensorFlow Roadmap](https://www.tensorflow.org/model_optimization/guide/roadmap)
+*   [TensorFlow model optimization roadmap](https://www.tensorflow.org/model_optimization/guide/roadmap)
 *   [TensorFlow White Papers](https://www.tensorflow.org/about/bib)
 *   [TensorBoard Visualization Toolkit](https://github.com/tensorflow/tensorboard)
 

RELEASE.md

@@ -35,7 +35,16 @@
 *   TF Core:
     *   Added `tf.saved_model.experimental.TrackableResource`, which allows the
         creation of custom wrapper objects for resource tensors.
-
+    *   Added `tf.lookup.experimental.MutableHashTable`, which provides a
+        generic mutable hash table implementation.
+        *   Compared to `tf.lookup.experimental.DenseHashTable` this offers
+        lower overall memory usage, and a cleaner API. It does not require
+        specifying a `delete_key` and `empty_key` that cannot be inserted into
+        the table.
+*   `tf.data`:
+    *   Promoting `tf.data.experimental.get_single_element` API to
+        `tf.data.Dataset.get_single_element` and deprecating the experimental
+        endpoint.
 *  `tf.lite`:
   * Fix mean op reference quantization rounding issue.
 

tensorflow/c/eager/BUILD

@@ -672,6 +672,7 @@ tf_cuda_cc_test(
         ":c_api_test_util",
         ":tfe_op_internal",
         ":tfe_tensorhandle_internal",
+        "@com_google_absl//absl/strings",
         "//tensorflow/c:c_test_util",
         "//tensorflow/core:lib",
         "//tensorflow/core:lib_internal",
@@ -680,7 +681,9 @@ tf_cuda_cc_test(
         "//tensorflow/core:test_main",
         "//tensorflow/core/common_runtime/eager:eager_operation",
         "//tensorflow/core/common_runtime/eager:tensor_handle",
-        "@com_google_absl//absl/strings",
+        # copybara:uncomment_begin
+        # "@tf_runtime//backends/cpu:tf_ops_alwayslink",
+        # copybara:uncomment_end
     ],
 )
 

tensorflow/c/eager/gradient_checker.cc

@@ -50,11 +50,11 @@ Status RunAndMaybeSum(AbstractContext* ctx, Model forward,
                       absl::Span<AbstractTensorHandle* const> inputs,
                       absl::Span<AbstractTensorHandle*> outputs,
                       bool use_function) {
-  std::vector<AbstractTensorHandle*> model_outputs(1);
+  AbstractTensorHandle* model_outputs[1];
 
   // Run the model.
-  TF_RETURN_IF_ERROR(RunModel(forward, ctx, inputs,
-                              absl::MakeSpan(model_outputs), use_function));
+  TF_RETURN_IF_ERROR(
+      RunModel(forward, ctx, inputs, model_outputs, use_function));
   AbstractTensorHandlePtr model_out(model_outputs[0]);
 
   TF_Tensor* model_out_tensor;
@@ -83,8 +83,8 @@ Status RunAndMaybeSum(AbstractContext* ctx, Model forward,
   }
 
   // Reduce sum the output on all dimensions.
-  TF_RETURN_IF_ERROR(
-      ops::Sum(ctx, model_out.get(), sum_dims.get(), outputs, "sum_output"));
+  TF_RETURN_IF_ERROR(ops::Sum(ctx, model_out.get(), sum_dims.get(), &outputs[0],
+                              "sum_output"));
   return Status::OK();
 }
 // ========================= End Helper Functions==============================
@@ -122,7 +122,7 @@ Status CalcNumericalGrad(AbstractContext* ctx, Model forward,
   // Initialize auxilary data structures.
   vector<float> thetaPlus_data(num_elems);
   vector<float> thetaMinus_data(num_elems);
-  std::vector<AbstractTensorHandle*> f_outputs(1);
+  AbstractTensorHandle* f_outputs[1];
 
   // Numerical Grad Check
   for (int i = 0; i < num_elems; i++) {
@@ -164,25 +164,25 @@ Status CalcNumericalGrad(AbstractContext* ctx, Model forward,
 
     // Get f(theta + eps):
     theta_inputs[input_index] = thetaPlus.get();
-    TF_RETURN_IF_ERROR(RunAndMaybeSum(ctx, forward, theta_inputs,
-                                      absl::MakeSpan(f_outputs), use_function));
+    TF_RETURN_IF_ERROR(
+        RunAndMaybeSum(ctx, forward, theta_inputs, f_outputs, use_function));
     AbstractTensorHandlePtr fPlus(f_outputs[0]);
 
     // Get f(theta - eps):
     theta_inputs[input_index] = thetaMinus.get();
-    TF_RETURN_IF_ERROR(RunAndMaybeSum(ctx, forward, theta_inputs,
-                                      absl::MakeSpan(f_outputs), use_function));
+    TF_RETURN_IF_ERROR(
+        RunAndMaybeSum(ctx, forward, theta_inputs, f_outputs, use_function));
     AbstractTensorHandlePtr fMinus(f_outputs[0]);
 
     // Take Difference of both estimates: (f(theta + eps) - f(theta - eps)).
-    TF_RETURN_IF_ERROR(ops::Sub(ctx, fPlus.get(), fMinus.get(),
-                                absl::MakeSpan(f_outputs), "sub_top"));
+    TF_RETURN_IF_ERROR(
+        ops::Sub(ctx, fPlus.get(), fMinus.get(), f_outputs, "sub_top"));
     AbstractTensorHandlePtr fDiff(f_outputs[0]);
 
     // Calculate using the difference quotient definition:
     // (f(theta + eps) - f(theta - eps)) / (2 * eps).
-    TF_RETURN_IF_ERROR(ops::Div(ctx, fDiff.get(), two_eps.get(),
-                                absl::MakeSpan(f_outputs), "diff_quotient"));
+    TF_RETURN_IF_ERROR(
+        ops::Div(ctx, fDiff.get(), two_eps.get(), f_outputs, "diff_quotient"));
     AbstractTensorHandlePtr diff_quotient(f_outputs[0]);
 
     TF_Tensor* grad_tensor;

tensorflow/c/eager/gradient_checker_test.cc

@@ -65,15 +65,15 @@ void CompareNumericalAndManualGradients(
 Status MatMulModel(AbstractContext* ctx,
                    absl::Span<AbstractTensorHandle* const> inputs,
                    absl::Span<AbstractTensorHandle*> outputs) {
-  return ops::MatMul(ctx, inputs[0], inputs[1], outputs, "MatMul",
+  return ops::MatMul(ctx, inputs[0], inputs[1], &outputs[0], "MatMul",
                      /*transpose_a=*/false,
                      /*transpose_b=*/false);
 }
 
 Status MulModel(AbstractContext* ctx,
                 absl::Span<AbstractTensorHandle* const> inputs,
                 absl::Span<AbstractTensorHandle*> outputs) {
-  return ops::Mul(ctx, inputs[0], inputs[1], outputs, "Mul");
+  return ops::Mul(ctx, inputs[0], inputs[1], &outputs[0], "Mul");
 }
 
 // TODO(vnvo2409): Add more tests from `python/ops/gradient_checker_v2_test.py`.

tensorflow/c/eager/gradients_test.cc

@@ -119,11 +119,11 @@ Status RecordOperationWithNullGradientFunctionModel(
     absl::Span<AbstractTensorHandle*> outputs) {
   Tape tape(/*persistent=*/false);
   tape.Watch(inputs[0]);
-  std::vector<AbstractTensorHandle*> neg_outputs(1);
-  TF_RETURN_IF_ERROR(
-      ops::Neg(ctx, inputs[0], absl::MakeSpan(neg_outputs), "Neg"));
-  tape.RecordOperation(inputs, neg_outputs, nullptr, "Neg");
-  return tape.ComputeGradient(ctx, /*targets=*/neg_outputs,
+  AbstractTensorHandle* neg_output;
+  TF_RETURN_IF_ERROR(ops::Neg(ctx, inputs[0], &neg_output, "Neg"));
+  tape.RecordOperation(inputs, {neg_output}, nullptr, "Neg");
+  return tape.ComputeGradient(ctx,
+                              /*targets=*/{neg_output},
                               /*sources=*/inputs,
                               /*output_gradients=*/{}, outputs);
 }

tensorflow/c/experimental/gradients/custom_gradient_test.cc

@@ -70,20 +70,17 @@ Status ExpWithPassThroughGrad(AbstractContext* ctx,
                               absl::Span<AbstractTensorHandle*> outputs) {
   Tape tape(/*persistent=*/false);
   tape.Watch(inputs[0]);  // Watch x.
-  std::vector<AbstractTensorHandle*> exp_outputs(1);
-  TF_RETURN_IF_ERROR(
-      ops::Exp(ctx, inputs[0], absl::MakeSpan(exp_outputs), "Exp"));
+  AbstractTensorHandle* exp_output;
+  TF_RETURN_IF_ERROR(ops::Exp(ctx, inputs[0], &exp_output, "Exp"));
   std::unique_ptr<GradientFunction> gradient_function(
       new PassThroughGradientFunction);
-  tape.RecordOperation(inputs, exp_outputs, gradient_function.release());
+  tape.RecordOperation(inputs, {exp_output}, gradient_function.release());
   TF_RETURN_IF_ERROR(tape.ComputeGradient(ctx,
-                                          /*targets*/ exp_outputs,
+                                          /*targets*/ {exp_output},
                                           /*sources=*/inputs,
                                           /*output_gradients=*/{},
                                           /*result=*/outputs));
-  for (auto exp_output : exp_outputs) {
-    exp_output->Unref();
-  }
+  exp_output->Unref();
   return Status::OK();
 }