Add CI for Habana Labs HPU/Gaudi2

.github/workflows/hpu-gaudi2.yml

@@ -0,0 +1,48 @@
+name: hpu-gaudi2
+
+on:
+  workflow_dispatch:
+  schedule:
+    - cron: "0 0 * * *"
+  pull_request:
+    paths:
+      - ".github/workflows/hpu-gaudi2.yml"
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+permissions:
+  contents: read
+  issues: write
+
+jobs:
+  unit-tests:
+    # The type of runner that the job will run on
+    runs-on: [self-hosted, intel, gaudi2]
+    container:
+      image: vault.habana.ai/gaudi-docker/1.14.0/ubuntu22.04/habanalabs/pytorch-installer-2.1.1:latest
+      ports:
+        - 80
+      options: --runtime=habana -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice
+
+    # Steps represent a sequence of tasks that will be executed as part of the job
+    steps:
+      # Checks-out your repository under $GITHUB_WORKSPACE, so your job can access it
+      - uses: actions/checkout@v3
+
+      - name: Check container state
+        run: |
+          ldd --version
+          hl-smi
+          python -c "import torch; print('torch:', torch.__version__, torch)"
+          python -c "import torch; print('CUDA available:', torch.cuda.is_available())"
+
+      - name: Install deepspeed
+        run: |
+          pip install .[dev]
+          ds_report
+
+      - name: Python environment
+        run: |
+          pip list

tests/unit/util.py

@@ -47,11 +47,14 @@ def bf16_required_version_check(accelerator_check=True):
     cuda_version_available = CUDA_MAJOR >= 11
     nccl_version_available = NCCL_MAJOR > 2 or (NCCL_MAJOR == 2 and NCCL_MINOR >= 10)
     npu_available = get_accelerator().device_name() == 'npu'
+    hpu_available = get_accelerator().device_name() == 'hpu'
 
     if torch_version_available and cuda_version_available and nccl_version_available and accelerator_pass:
         return True
     elif npu_available:
         return True
+    elif hpu_available:
+        return True
     else:
         return False