Create fork-maintenance.yml (#61)

Add fork maintenance system and add skip for multigpu issue

.github/workflows/fork-maintenance.yml

@@ -0,0 +1,30 @@
+name: Run Scheduled Events Docker
+
+permissions:
+  actions: write
+  contents: write
+  issues: write
+  pull-requests: write
+
+on:
+  workflow_dispatch:
+  schedule:
+    - cron: '0 0 10 * *'
+
+jobs:
+  run-scheduled-events:
+    uses: Cemberk/fork-maintenance-system/.github/workflows/fork-maintenance-action.yml@artifacts
+    with:
+      platform: 'gfx90a'
+      upstream_repo: 'https://github.com/huggingface/transformers'
+      pr_branch_prefix: 'scheduled-merge'
+      requirements_command: 'sudo sed -i \"s/torchaudio//g\" examples/pytorch/_tests_requirements.txt && pip install -r examples/pytorch/_tests_requirements.txt && git restore examples/pytorch/_tests_requirements.txt && pip install --no-cache-dir GPUtil azureml azureml-core tokenizers ninja cerberus sympy sacremoses sacrebleu==1.5.1 sentencepiece scipy scikit-learn urllib3 && pip install huggingface_hub datasets && pip install parameterized && pip install -e .'
+      unit_test_command: cd tests; folders=\$(python3 -c \import os; tests = os.getcwd(); models = \"models\";  model_tests = os.listdir(os.path.join(tests, models)); d1 = sorted(list(filter(os.path.isdir, os.listdir(tests)))); d2 = sorted(list(filter(os.path.isdir, [os.path.join(models, x) for x in model_tests]))); d1.remove(models); d = d2 + d1; print(\" \".join(d[:5]))' ); cd ..; for folder in \${folders[@]}; do pytest tests/\${folder} -v --make-reports=huggingface_unit_tests_\${machine_type}_run_models_gpu_\${folder} -rfEs --continue-on-collection-errors -m \"not not_device_test\" -p no:cacheprovider; done;  allstats=\$(find reports -name stats.txt);  for stat in \${allstats[@]}; do echo \$stat; cat \$stat; done
+      performance_test_command: 'echo \"python examples/pytorch/language-modeling/run_mlm.py --model_name_or_path bert-base-uncased --dataset_name wikitext --dataset_config_name wikitext-2-raw-v1 --do_train --do_eval --output_dir /tmp/test-mlm --per_device_train_batch_size 8 --per_device_eval_batch_size 8 --max_steps 500\"'
+      docker_image: 'rocm/pytorch:latest'
+      docker_options: '--device=/dev/kfd --device=/dev/dri --group-add video --shm-size 16G --network=host'
+    secrets:
+      GIT_TOKEN: ${{ secrets.CRED_TOKEN }}
+      schedule_json: ${{ secrets.SCHEDULE_CONFIG }}
+
+

tests/test_modeling_common.py

@@ -3067,6 +3067,7 @@ def test_inputs_embeds_matches_input_ids_with_generate(self):
             self.assertTrue(number_of_different_tokens <= max_differences)  # accept up to 10% mismatch
 
     @require_non_xpu
+    @skipIfRocm
     @require_torch_multi_gpu
     def test_multi_gpu_data_parallel_forward(self):
         config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()