Merge branch 'master' of https://github.com/openvinotoolkit/openvino …

…into as/weights_bank_dq_update
openvinotoolkit · Oct 3, 2024 · d82f570 · d82f570
2 parents 52fc4cf + 2fc0fae
commit d82f570
Show file tree

Hide file tree

Showing 26 changed files with 1,750 additions and 94 deletions.
diff --git a/.github/workflows/job_pytorch_layer_tests.yml b/.github/workflows/job_pytorch_layer_tests.yml
@@ -7,10 +7,6 @@ on:
         description: 'Machine on which the tests would run'
         type: string
         required: true
-      shell:
-        description: "shell to override the default shell settings in the runner's operating system."
-        type: string
-        required: true
       container:
         description: 'JSON to be converted to the value of the "container" configuration for the job'
         type: string
@@ -20,12 +16,15 @@ on:
         description: 'Components that are affected by changes in the commit defined by the Smart CI Action'
         type: string
         required: true
+      python-version:
+        description: 'Python version to setup. E.g., "3.11"'
+        type: string
+        required: true
 
 permissions: read-all
 
 env:
   PIP_CACHE_PATH: /mount/caches/pip/linux
-  PYTHON_VERSION: '3.11'
 
 jobs:
   PyTorch_Layer_Tests:
@@ -35,7 +34,7 @@ jobs:
     container: ${{ fromJSON(inputs.container) }}
     defaults:
       run:
-        shell: ${{ inputs.shell }}
+        shell: ${{ contains(inputs.runner, 'win') && 'pwsh' || 'bash' }}
     env:
       DEBIAN_FRONTEND: noninteractive # to prevent apt-get from waiting user input
       OPENVINO_REPO: ${{ github.workspace }}/openvino
@@ -55,12 +54,6 @@ jobs:
           name: openvino_tests
           path: ${{ env.INSTALL_TEST_DIR }}
 
-      - name: Download OpenVINO tokenizers extension
-        uses: actions/download-artifact@fa0a91b85d4f404e444e00e005971372dc801d16 # v4.1.8
-        with:
-          name: openvino_tokenizers_wheel
-          path: ${{ env.INSTALL_DIR }}
-
       # Needed as ${{ github.workspace }} is not working correctly when using Docker
       - name: Setup Variables
         if: runner.os != 'Windows'
@@ -98,10 +91,10 @@ jobs:
           sparse-checkout-cone-mode: false
           path: 'openvino'
 
-      - name: Setup Python ${{ env.PYTHON_VERSION }}
+      - name: Setup Python ${{ inputs.python-version }}
         uses: ./openvino/.github/actions/setup_python
         with:
-          version: ${{ env.PYTHON_VERSION }}
+          version: ${{ inputs.python-version }}
           pip-cache-path: ${{ runner.os == 'Linux' && env.PIP_CACHE_PATH || '' }}
           should-setup-pip-paths: ${{ runner.os == 'Linux' }}
           self-hosted-runner: ${{ runner.os == 'Linux' }}
@@ -112,43 +105,39 @@ jobs:
           # Install the core OV wheel
           python3 -m pip install ${INSTALL_DIR}/tools/openvino-*.whl
 
-          # Install the core OV Tokenizers wheel
-          python3 -m pip install ${INSTALL_DIR}/openvino_tokenizers-*.whl
-
       - name: Install OpenVINO Python wheels (Windows)
         if: runner.os == 'Windows'
         run: |
           # Find and install the core OV wheel
           $ovCoreWheelPath=Get-ChildItem -Path ${{ env.INSTALL_DIR }}\tools -Filter openvino-*.whl | % { $_.FullName }
           python3 -m pip install "$ovCoreWheelPath"
 
-          # Find and install the core OV Tokenizers wheel
-          $ovCoreWheelPath=Get-ChildItem -Path ${{ env.INSTALL_DIR }} -Filter openvino_tokenizers-*.whl | % { $_.FullName }
-          python3 -m pip install "$ovCoreWheelPath"
-
       - name: Install Pytorch Layer tests dependencies
         run: |
           # pytorch test requirements
           python3 -m pip install -r ${{ env.INSTALL_TEST_DIR }}/requirements_pytorch
 
       - name: PyTorch Layer Tests
         if: ${{ fromJSON(inputs.affected-components).PyTorch_FE.test && runner.arch != 'ARM64' }} # Ticket: 126287, 142196
-        run: python3 -m pytest ${{ env.LAYER_TESTS_INSTALL_DIR }}/pytorch_tests -n logical -m precommit --junitxml=${{ env.INSTALL_TEST_DIR }}/TEST-pytorch.xml
+        # due to CVS-152795, parallel run is not possible on Windows
+        run: python3 -m pytest ${{ env.LAYER_TESTS_INSTALL_DIR }}/pytorch_tests ${PARALLEL} -m precommit --junitxml=${{ env.INSTALL_TEST_DIR }}/TEST-pytorch.xml
         env:
           TEST_DEVICE: CPU
           TEST_PRECISION: FP32
+          PARALLEL: ${{ runner.os == 'Windows' && ' ' || '-n logical'}}
 
       - name: PyTorch torch.export Layer Tests
-        if: ${{ fromJSON(inputs.affected-components).PyTorch_FE.test && runner.arch != 'ARM64' }} # Ticket: 126287
+        if: ${{ fromJSON(inputs.affected-components).PyTorch_FE.test && runner.arch != 'ARM64' && runner.os != 'Windows' }} # Ticket: 126287
         run: |
-          python3 -m pytest ${{ env.LAYER_TESTS_INSTALL_DIR }}/pytorch_tests -n logical -m precommit_torch_export --junitxml=${{ env.INSTALL_TEST_DIR }}/TEST-pytorch.xml
+          python3 -m pytest ${{ env.LAYER_TESTS_INSTALL_DIR }}/pytorch_tests ${PARALLEL} -m precommit_torch_export --junitxml=${{ env.INSTALL_TEST_DIR }}/TEST-pytorch.xml
         env:
           TEST_DEVICE: CPU
           TEST_PRECISION: FP32
           PYTORCH_TRACING_MODE: EXPORT
+          PARALLEL: ${{ runner.os == 'Windows' && ' ' || '-n logical'}}
 
       - name: PyTorch torch.compile TORCHFX Layer Tests
-        if: ${{ fromJSON(inputs.affected-components).PyTorch_FE.test && runner.os != 'macOS' && runner.arch != 'ARM64' }} # Ticket: 126287
+        if: ${{ fromJSON(inputs.affected-components).PyTorch_FE.test && runner.os != 'macOS' && runner.arch != 'ARM64' && runner.os != 'Windows' }} # Ticket: 126287
         run: |
           python3 -m pytest ${{ env.LAYER_TESTS_INSTALL_DIR }}/pytorch_tests -m precommit_fx_backend --junitxml=${{ env.INSTALL_TEST_DIR }}/TEST-pytorch.xml
         env:

diff --git a/.github/workflows/job_tensorflow_layer_tests.yml b/.github/workflows/job_tensorflow_layer_tests.yml
@@ -7,10 +7,6 @@ on:
         description: 'Machine on which the tests would run'
         type: string
         required: true
-      shell:
-        description: "shell to override the default shell settings in the runner's operating system."
-        type: string
-        required: true
       container:
         description: 'JSON to be converted to the value of the "container" configuration for the job'
         type: string
@@ -20,12 +16,15 @@ on:
         description: 'Components that are affected by changes in the commit defined by the Smart CI Action'
         type: string
         required: true
+      python-version:
+        description: 'Python version to setup. E.g., "3.11"'
+        type: string
+        required: true
 
 permissions: read-all
 
 env:
   PIP_CACHE_PATH: /mount/caches/pip/linux
-  PYTHON_VERSION: '3.11'
 
 jobs:
   TensorFlow_Layer_Tests:
@@ -35,7 +34,7 @@ jobs:
     container: ${{ fromJSON(inputs.container) }}
     defaults:
       run:
-        shell: ${{ inputs.shell }}
+        shell: ${{ contains(inputs.runner, 'win') && 'pwsh' || 'bash' }}
     env:
       DEBIAN_FRONTEND: noninteractive # to prevent apt-get from waiting user input
       OPENVINO_REPO: ${{ github.workspace }}/openvino
@@ -98,10 +97,10 @@ jobs:
           sparse-checkout-cone-mode: false
           path: 'openvino'
 
-      - name: Setup Python ${{ env.PYTHON_VERSION }}
+      - name: Setup Python ${{ inputs.python-version }}
         uses: ./openvino/.github/actions/setup_python
         with:
-          version: ${{ env.PYTHON_VERSION }}
+          version: ${{ inputs.python-version }}
           pip-cache-path: ${{ runner.os == 'Linux' && env.PIP_CACHE_PATH || '' }}
           should-setup-pip-paths: ${{ runner.os == 'Linux' }}
           self-hosted-runner: ${{ runner.os == 'Linux' }}

diff --git a/.github/workflows/linux_arm64.yml b/.github/workflows/linux_arm64.yml
@@ -173,19 +173,19 @@ jobs:
     uses: ./.github/workflows/job_tensorflow_layer_tests.yml
     with:
       runner: 'aks-linux-16-cores-arm'
-      shell: bash
       container: '{"image": "${{ fromJSON(needs.docker.outputs.images).ov_test.ubuntu_20_04_arm64 }}", "volumes": ["/mount:/mount"]}'
       affected-components: ${{ needs.smart_ci.outputs.affected_components }}
+      python-version: '3.11'
 
   Pytorch_Layer_Tests:
     name: Pytorch Layer Tests
-    needs: [ Build, Docker, Smart_CI, Openvino_tokenizers ]
+    needs: [ Build, Docker, Smart_CI ]
     uses: ./.github/workflows/job_pytorch_layer_tests.yml
     with:
       runner: 'aks-linux-16-cores-arm'
-      shell: bash
       container: '{"image": "${{ fromJSON(needs.docker.outputs.images).ov_test.ubuntu_20_04_arm64 }}", "volumes": ["/mount:/mount"]}'
       affected-components: ${{ needs.smart_ci.outputs.affected_components }}
+      python-version: '3.11'
 
   CPU_Functional_Tests:
     name: CPU functional tests

diff --git a/.github/workflows/mac.yml b/.github/workflows/mac.yml
@@ -276,17 +276,17 @@ jobs:
     uses: ./.github/workflows/job_tensorflow_layer_tests.yml
     with:
       runner: 'macos-13'
-      shell: bash
       affected-components: ${{ needs.smart_ci.outputs.affected_components }}
+      python-version: '3.11'
 
   Pytorch_Layer_Tests:
     name: Pytorch Layer Tests
-    needs: [ Build, Smart_CI, Openvino_tokenizers ]
+    needs: [ Build, Smart_CI ]
     uses: ./.github/workflows/job_pytorch_layer_tests.yml
     with:
       runner: 'macos-13'
-      shell: bash
       affected-components: ${{ needs.smart_ci.outputs.affected_components }}
+      python-version: '3.11'
 
   CPU_Functional_Tests:
     name: CPU functional tests

diff --git a/.github/workflows/mac_arm64.yml b/.github/workflows/mac_arm64.yml
@@ -275,17 +275,17 @@ jobs:
     uses: ./.github/workflows/job_tensorflow_layer_tests.yml
     with:
       runner: 'macos-13-xlarge'
-      shell: bash
       affected-components: ${{ needs.smart_ci.outputs.affected_components }}
+      python-version: '3.11'
 
   Pytorch_Layer_Tests:
     name: Pytorch Layer Tests
-    needs: [ Build, Smart_CI, Openvino_tokenizers ]
+    needs: [ Build, Smart_CI ]
     uses: ./.github/workflows/job_pytorch_layer_tests.yml
     with:
       runner: 'macos-13-xlarge'
-      shell: bash
       affected-components: ${{ needs.smart_ci.outputs.affected_components }}
+      python-version: '3.11'
 
   CPU_Functional_Tests:
     name: CPU functional tests

diff --git a/.github/workflows/ubuntu_22.yml b/.github/workflows/ubuntu_22.yml
@@ -305,19 +305,19 @@ jobs:
     uses: ./.github/workflows/job_tensorflow_layer_tests.yml
     with:
       runner: 'aks-linux-4-cores-16gb'
-      shell: bash
       container: '{"image": "${{ fromJSON(needs.docker.outputs.images).ov_test.ubuntu_22_04_x64 }}", "volumes": ["/mount:/mount"]}'
       affected-components: ${{ needs.smart_ci.outputs.affected_components }}
+      python-version: '3.11'
 
   Pytorch_Layer_Tests:
     name: Pytorch Layer Tests
-    needs: [ Docker, Build, Smart_CI, Openvino_tokenizers ]
+    needs: [ Docker, Build, Smart_CI ]
     uses: ./.github/workflows/job_pytorch_layer_tests.yml
     with:
       runner: 'aks-linux-4-cores-16gb'
-      shell: bash
       container: '{"image": "${{ fromJSON(needs.docker.outputs.images).ov_test.ubuntu_22_04_x64 }}", "volumes": ["/mount:/mount"]}'
       affected-components: ${{ needs.smart_ci.outputs.affected_components }}
+      python-version: '3.11'
 
   CPU_Functional_Tests:
     name: CPU functional tests

diff --git a/.github/workflows/ubuntu_24.yml b/.github/workflows/ubuntu_24.yml
@@ -133,6 +133,16 @@ jobs:
       affected-components: ${{ needs.smart_ci.outputs.affected_components }}
       python-version: '3.12'
 
+  Pytorch_Layer_Tests:
+    name: Pytorch Layer Tests
+    needs: [ Docker, Build, Smart_CI ]
+    uses: ./.github/workflows/job_pytorch_layer_tests.yml
+    with:
+      runner: 'aks-linux-4-cores-16gb'
+      container: '{"image": "${{ fromJSON(needs.docker.outputs.images).ov_test.ubuntu_24_04_x64 }}", "volumes": ["/mount:/mount"]}'
+      affected-components: ${{ needs.smart_ci.outputs.affected_components }}
+      python-version: '3.12'
+
   Overall_Status:
     name: ci/gha_overall_status_ubuntu_24
     needs: [Smart_CI, Build, Debian_Packages, Samples, Python_Unit_Tests]

diff --git a/.github/workflows/windows_vs2019_release.yml b/.github/workflows/windows_vs2019_release.yml
@@ -404,17 +404,17 @@ jobs:
     uses: ./.github/workflows/job_tensorflow_layer_tests.yml
     with:
       runner: 'aks-win-8-cores-16gb'
-      shell: pwsh
       affected-components: ${{ needs.smart_ci.outputs.affected_components }}
+      python-version: '3.11'
 
   Pytorch_Layer_Tests:
     name: Pytorch Layer Tests
-    needs: [ Build, Smart_CI, Openvino_tokenizers ]
+    needs: [ Build, Smart_CI ]
     uses: ./.github/workflows/job_pytorch_layer_tests.yml
     with:
       runner: 'aks-win-8-cores-16gb'
-      shell: pwsh
       affected-components: ${{ needs.smart_ci.outputs.affected_components }}
+      python-version: '3.11'
 
   CXX_Unit_Tests:
     name: C++ unit tests

diff --git a/.../openvino-workflow/running-inference/inference-devices-and-modes/npu-device.rst b/.../openvino-workflow/running-inference/inference-devices-and-modes/npu-device.rst
@@ -11,6 +11,7 @@ NPU Device
    :hidden:
 
    npu-device/remote-tensor-api-npu-plugin
+   npu-device/batching-on-npu-plugin
 
 
 The Neural Processing Unit is a low-power hardware solution, introduced with the

diff --git a/...ing-inference/inference-devices-and-modes/npu-device/batching-on-npu-plugin.rst b/...ing-inference/inference-devices-and-modes/npu-device/batching-on-npu-plugin.rst
@@ -0,0 +1,37 @@
+NPU Plugin Batching 
+===============================
+
+
+.. meta::
+   :description: OpenVINO™ NPU plugin supports batching
+                 either by executing concurrent inferences or by
+                 relying on native compiler support for batching.
+
+OpenVINO™ NPU plugin supports batching either by executing concurrent inferences or by relying on native compiler support for batching.
+
+First, the NPU plugin checks if the following conditions are met:
+
+* The batch size is on the first axis.
+* All inputs and outputs have the same batch size.
+* The model does not contain states.
+
+**If the conditions are met**, the NPU plugin attempts to compile and execute the original model with batch_size forced to 1. This approach is due to current compiler limitations and ongoing work to improve performance for batch_size greater than one.
+If the compilation is successful, the plugin detects a difference in batch size between the original model layout (with a batch size set to N)
+and the transformed/compiled layout (with a batch size set to 1). Then it executes the following steps:
+
+1. Internally constructs multiple command lists, one for each input.
+2. Executes each command list for the proper offsets of input/output buffers.
+3. Notifies the user of the completion of the inference request after all command lists have been executed.
+
+This concurrency-based batching mode is transparent to the application. A single inference request handles all inputs from the batch.
+While performance may be lower compared to regular batching (based on native compiler support), this mode provides basic batching functionality for use either with older drivers
+or when the model cannot yet be compiled with a batch size larger than one.
+
+**If the conditions are not met**, the NPU plugin tries to compile and execute the original model with the given
+batch_size to N as any other regular model.
+
+.. note::
+
+   With future performance improvements and support for compiling multiple models with a batch size larger 
+   than one, the default order will change. NPU will try first to compile and execute the original model with the 
+   given batch size and fall back to concurrent batching if compilation fails.
diff --git a/src/frontends/tensorflow_common/src/op/expand_dims.cpp b/src/frontends/tensorflow_common/src/op/expand_dims.cpp
@@ -3,7 +3,13 @@
 //
 
 #include "common_op_table.hpp"
+#include "helper_ops/complex_type_mark.hpp"
+#include "openvino/op/constant.hpp"
+#include "openvino/op/less.hpp"
+#include "openvino/op/select.hpp"
+#include "openvino/op/subtract.hpp"
 #include "openvino/op/unsqueeze.hpp"
+#include "utils.hpp"
 
 using namespace std;
 using namespace ov::op;
@@ -14,9 +20,31 @@ namespace tensorflow {
 namespace op {
 
 OutputVector translate_expand_dims_op(const NodeContext& node) {
-    default_op_checks(node, 2, {"ExpandDims", "EXPAND_DIMS"});
+    default_op_checks(node, 2, {"ExpandDims", "EXPAND_DIMS"}, true);
     auto input = node.get_input(0);
     auto axis = node.get_input(1);
+    auto complex_type_mark = as_type_ptr<ComplexTypeMark>(input.get_node_shared_ptr());
+
+    if (complex_type_mark) {
+        element::Type complex_part_type = complex_type_mark->get_complex_part_type();
+        input = complex_type_mark->input_value(0);
+
+        auto const_zero = create_same_type_const_scalar<int32_t>(axis, 0);
+
+        auto is_axis_neg = make_shared<v1::Less>(axis, const_zero);
+
+        auto const_one = create_same_type_const_scalar<int32_t>(axis, 1);
+        auto axis_min_one = make_shared<v1::Subtract>(axis, const_one);
+
+        auto new_axis = make_shared<v1::Select>(is_axis_neg, axis_min_one, axis);
+
+        auto unsqueeze = make_shared<v0::Unsqueeze>(input, new_axis);
+
+        set_node_name(node.get_name(), unsqueeze);
+        auto complex_result = make_shared<ComplexTypeMark>(unsqueeze, complex_part_type);
+        return {complex_result};
+    }
+
     auto unsqueeze = make_shared<v0::Unsqueeze>(input, axis);
     set_node_name(node.get_name(), unsqueeze);
     return {unsqueeze};