Fix many HPU failures in the CI (#39066)

* more torch.hpu patches

* increase top_k because it results in flaky behavior when Tempreture, TopP and TopK are used together, which ends up killing beams early.

* remove temporal fix

* fix scatter operation when input and src are the same

* trigger

* fix and reduce

* skip finding batch size as it makes the hpu go loco

* fix fsdp (yay all are passing)

* fix checking equal nan values

* style

* remove models list

* order

* rename to cuda_extensions

* Update src/transformers/trainer.py

.github/workflows/self-scheduled-intel-gaudi.yml

@@ -84,8 +84,6 @@ jobs:
       machine_type: ${{ matrix.machine_type }}
       folder_slices: ${{ needs.setup.outputs.folder_slices }}
       runner: ${{ inputs.runner_scale_set }}-${{ matrix.machine_type }}
-      report_name_prefix: run_models_gpu
-
     secrets: inherit
 
   run_trainer_and_fsdp_gpu:
@@ -104,11 +102,10 @@ jobs:
       folder_slices: ${{ needs.setup.outputs.folder_slices }}
       runner: ${{ inputs.runner_scale_set }}-${{ matrix.machine_type }}
       report_name_prefix: run_trainer_and_fsdp_gpu
-
     secrets: inherit
 
-  run_pipelines_gpu:
-    if: ${{ inputs.job == 'run_pipelines_gpu' }}
+  run_pipelines_torch_gpu:
+    if: ${{ inputs.job == 'run_pipelines_torch_gpu' }}
     name: Pipelines
     strategy:
       fail-fast: false
@@ -161,20 +158,20 @@ jobs:
 
       - name: Run all pipeline tests on Intel Gaudi
         run: |
-          python3 -m pytest -v --make-reports=${{ env.machine_type }}_run_pipelines_gpu_test_reports tests/pipelines -m "not not_device_test"
+          python3 -m pytest -v --make-reports=${{ env.machine_type }}_run_pipelines_torch_gpu_test_reports tests/pipelines -m "not not_device_test"
 
       - name: Failure short reports
         if: ${{ failure() }}
         continue-on-error: true
         run: |
-          cat reports/${{ env.machine_type }}_run_pipelines_gpu_test_reports/failures_short.txt
+          cat reports/${{ env.machine_type }}_run_pipelines_torch_gpu_test_reports/failures_short.txt
 
-      - name: "Test suite reports artifacts: ${{ env.machine_type }}_run_pipelines_gpu_test_reports"
+      - name: "Test suite reports artifacts: ${{ env.machine_type }}_run_pipelines_torch_gpu_test_reports"
         if: ${{ always() }}
         uses: actions/upload-artifact@v4
         with:
-          name: ${{ env.machine_type }}_run_pipelines_gpu_test_reports
-          path: reports/${{ env.machine_type }}_run_pipelines_gpu_test_reports
+          name: ${{ env.machine_type }}_run_pipelines_torch_gpu_test_reports
+          path: reports/${{ env.machine_type }}_run_pipelines_torch_gpu_test_reports
 
   run_examples_gpu:
     if: ${{ inputs.job == 'run_examples_gpu' }}
@@ -248,8 +245,8 @@ jobs:
           name: ${{ env.machine_type }}_run_examples_gpu_test_reports
           path: reports/${{ env.machine_type }}_run_examples_gpu_test_reports
 
-  run_deepspeed_gpu:
-    if: ${{ inputs.job == 'run_deepspeed_gpu' }}
+  run_torch_cuda_extensions_gpu:
+    if: ${{ inputs.job == 'run_torch_cuda_extensions_gpu' }}
     name: Intel Gaudi deepspeed tests
     strategy:
       fail-fast: false
@@ -305,20 +302,20 @@ jobs:
 
       - name: Run all deepspeed tests on intel Gaudi
         run: |
-          python3 -m pytest -v --make-reports=${{ env.machine_type }}_run_deepspeed_gpu_test_reports tests/deepspeed -m "not not_device_test"
+          python3 -m pytest -v --make-reports=${{ env.machine_type }}_run_torch_cuda_extensions_gpu_test_reports tests/deepspeed -m "not not_device_test"
 
       - name: Failure short reports
         if: ${{ failure() }}
         continue-on-error: true
         run: |
-          cat reports/${{ env.machine_type }}_run_deepspeed_gpu_test_reports/failures_short.txt
+          cat reports/${{ env.machine_type }}_run_torch_cuda_extensions_gpu_test_reports/failures_short.txt
 
-      - name: "Test suite reports artifacts: ${{ env.machine_type }}_run_deepspeed_gpu_test_reports"
+      - name: "Test suite reports artifacts: ${{ env.machine_type }}_run_torch_cuda_extensions_gpu_test_reports"
         if: ${{ always() }}
         uses: actions/upload-artifact@v4
         with:
-          name: ${{ env.machine_type }}_run_deepspeed_gpu_test_reports
-          path: reports/${{ env.machine_type }}_run_deepspeed_gpu_test_reports
+          name: ${{ env.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
+          path: reports/${{ env.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
 
   send_results:
     name: Slack Report
@@ -327,8 +324,8 @@ jobs:
         setup,
         run_models_gpu,
         run_examples_gpu,
-        run_pipelines_gpu,
-        run_deepspeed_gpu,
+        run_torch_cuda_extensions_gpu,
+        run_pipelines_torch_gpu,
         run_trainer_and_fsdp_gpu,
       ]
     if: ${{ always() }}

.github/workflows/self-scheduled-intel-gaudi3-caller.yml

@@ -23,7 +23,7 @@ jobs:
     name: Pipeline CI
     uses: ./.github/workflows/self-scheduled-intel-gaudi.yml
     with:
-      job: run_pipelines_gpu
+      job: run_pipelines_torch_gpu
       ci_event: Scheduled CI (Intel) - Gaudi3
       runner_scale_set: itac-bm-emr-gaudi3-dell
       slack_report_channel: "#transformers-ci-daily-intel-gaudi3"
@@ -47,7 +47,7 @@ jobs:
     name: DeepSpeed CI
     uses: ./.github/workflows/self-scheduled-intel-gaudi.yml
     with:
-      job: run_deepspeed_gpu
+      job: run_torch_cuda_extensions_gpu
       ci_event: Scheduled CI (Intel) - Gaudi3
       runner_scale_set: itac-bm-emr-gaudi3-dell
       slack_report_channel: "#transformers-ci-daily-intel-gaudi3"