test

.github/workflows/upstream_tests.yml

@@ -0,0 +1,151 @@
+name: Upstream Integrations
+
+on:
+  workflow_dispatch:
+  schedule:
+    - cron: "0 0 * * *" # Run daily at 00:00 UTC
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
+  cancel-in-progress: true
+
+jobs:
+  transformers:
+    name: Upstream Transformers
+    runs-on:
+      group: aws-dl1-24xlarge
+
+    container:
+      image: docker://vault.habana.ai/gaudi-docker/1.20.0/ubuntu22.04/habanalabs/pytorch-installer-2.6.0:latest
+      options: --runtime=habana --shm-size=64G --env HABANA_VISIBLE_DEVICES
+      env:
+        OMPI_MCA_btl_vader_single_copy_mechanism: none
+        RUN_THIRD_PARTY_DEVICE_TESTS: 1
+        TRANSFORMERS_TEST_DEVICE: hpu
+        PT_ENABLE_INT64_SUPPORT: 1
+        PT_HPU_LAZY_MODE: 0
+        RUN_SLOW: 1
+
+    steps:
+      - name: HL-SMI (1)
+        run: |
+          hl-smi
+          echo "HABANA_VISIBLE_DEVICES=${HABANA_VISIBLE_DEVICES}"
+          echo "HABANA_VISIBLE_MODULES=${HABANA_VISIBLE_MODULES}"
+
+      - name: Extract HPU visible modules
+        id: add-modules
+        run: |
+          export HABANA_VISIBLE_MODULES=$(hl-smi -Q module_id -f csv,noheader | tr '\n' ',' | sed 's/,$//')
+          echo "HABANA_VISIBLE_MODULES=${HABANA_VISIBLE_MODULES}" >> $GITHUB_ENV
+
+      - name: HL-SMI (2)
+        run: |
+          hl-smi
+          echo "HABANA_VISIBLE_DEVICES=${HABANA_VISIBLE_DEVICES}"
+          echo "HABANA_VISIBLE_MODULES=${HABANA_VISIBLE_MODULES}"
+
+      - name: Checkout to Transformers
+        uses: actions/checkout@v4
+        with:
+          ref: hpu-support
+          repository: huggingface/transformers
+
+      - name: Install Transformers with Accelerate and DeepSpeed
+        run: |
+          pip install -e .[testing] "numpy<2.0.0" scipy scikit-learn \
+            git+https://github.com/huggingface/accelerate.git@hpu-support \
+            git+https://github.com/HabanaAI/DeepSpeed.git@1.20.0
+
+      - name: Run Trainer tests
+        run: |
+          pytest tests/trainer/test_trainer.py -s -vvvv
+
+      - name: Run Trainer Utils tests
+        run: |
+          pytest tests/trainer/test_trainer_utils.py -s -vvvv
+
+      - name: Run Trainer Seq2Seq tests
+        run: |
+          pytest tests/trainer/test_trainer_seq2seq.py -s -vvvv
+
+      - name: Run Trainer Distributed tests
+        run: |
+          pytest tests/trainer/test_trainer_distributed.py -s -vvvv
+
+      - name: Run FSDP Integration tests
+        run: |
+          pytest tests/fsdp/test_fsdp.py tests/trainer/test_trainer_fsdp.py -s -vvvv
+
+      - name: Run DeepSpeed Integration tests
+        run: |
+          pytest tests/deepspeed/test_deepspeed.py -s -vvvv
+
+  accelerate:
+    name: Upstream Accelerate
+    runs-on:
+      group: aws-dl1-24xlarge
+
+    container:
+      image: docker://vault.habana.ai/gaudi-docker/1.20.0/ubuntu22.04/habanalabs/pytorch-installer-2.6.0:latest
+      options: --runtime=habana --shm-size=64G --cap-add=sys_nice --env HABANA_VISIBLE_DEVICES
+      env:
+        OMPI_MCA_btl_vader_single_copy_mechanism: none
+        PT_ENABLE_INT64_SUPPORT: 1
+        PT_HPU_LAZY_MODE: 0
+        RUN_SLOW: 1
+
+    steps:
+      - name: HL-SMI (1)
+        run: |
+          hl-smi
+          echo "HABANA_VISIBLE_DEVICES=${HABANA_VISIBLE_DEVICES}"
+          echo "HABANA_VISIBLE_MODULES=${HABANA_VISIBLE_MODULES}"
+
+      - name: Extract HPU visible modules
+        id: add-modules
+        run: |
+          export HABANA_VISIBLE_MODULES=$(hl-smi -Q module_id -f csv,noheader | tr '\n' ',' | sed 's/,$//')
+          echo "HABANA_VISIBLE_MODULES=${HABANA_VISIBLE_MODULES}" >> $GITHUB_ENV
+
+      - name: HL-SMI (2)
+        run: |
+          hl-smi
+          echo "HABANA_VISIBLE_DEVICES=${HABANA_VISIBLE_DEVICES}"
+          echo "HABANA_VISIBLE_MODULES=${HABANA_VISIBLE_MODULES}"
+
+      - name: Checkout to Accelerate
+        uses: actions/checkout@v4
+        with:
+          ref: hpu-support
+          repository: huggingface/accelerate
+
+      - name: Install Accelerate with Transformers and DeepSpeed
+        run: |
+          pip install -e .[testing] \
+            git+https://github.com/HabanaAI/DeepSpeed.git@1.20.0 \
+            git+https://github.com/huggingface/transformers.git@hpu-support
+
+      - name: Run CLI tests
+        run: |
+          make test_cli
+
+      - name: Run Core tests
+        run: |
+          make test_core
+
+      - name: Run Big Modeling tests
+        run: |
+          make test_big_modeling
+
+      - name: Run FSDP integration tests
+        run: |
+          make test_fsdp
+
+      - name: Run DeepSpeed integration tests
+        run: |
+          make test_deepspeed
+
+      - name: Run Examples tests
+        run: |
+          make test_examples