fix gptq tests

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

@@ -7,64 +7,64 @@ on:
     - cron: "17 2 * * *"
   push:
     branches:
-      - run_scheduled_ci*
+      - fix_gptq_test
 
 jobs:
-  model-ci:
-    name: Model CI
-    uses: ./.github/workflows/self-scheduled.yml
-    with:
-      job: run_models_gpu
-      slack_report_channel: "#transformers-ci-daily-models"
-      runner: daily-ci
-      docker: huggingface/transformers-all-latest-gpu
-      ci_event: Daily CI
-    secrets: inherit
+  # model-ci:
+  #   name: Model CI
+  #   uses: ./.github/workflows/self-scheduled.yml
+  #   with:
+  #     job: run_models_gpu
+  #     slack_report_channel: "#transformers-ci-daily-models"
+  #     runner: daily-ci
+  #     docker: huggingface/transformers-all-latest-gpu
+  #     ci_event: Daily CI
+  #   secrets: inherit
 
-  torch-pipeline:
-    name: Torch pipeline CI
-    uses: ./.github/workflows/self-scheduled.yml
-    with:
-      job: run_pipelines_torch_gpu
-      slack_report_channel: "#transformers-ci-daily-pipeline-torch"
-      runner: daily-ci
-      docker: huggingface/transformers-pytorch-gpu
-      ci_event: Daily CI
-    secrets: inherit
+  # torch-pipeline:
+  #   name: Torch pipeline CI
+  #   uses: ./.github/workflows/self-scheduled.yml
+  #   with:
+  #     job: run_pipelines_torch_gpu
+  #     slack_report_channel: "#transformers-ci-daily-pipeline-torch"
+  #     runner: daily-ci
+  #     docker: huggingface/transformers-pytorch-gpu
+  #     ci_event: Daily CI
+  #   secrets: inherit
 
-  tf-pipeline:
-    name: TF pipeline CI
-    uses: ./.github/workflows/self-scheduled.yml
-    with:
-      job: run_pipelines_tf_gpu
-      slack_report_channel: "#transformers-ci-daily-pipeline-tf"
-      runner: daily-ci
-      docker: huggingface/transformers-tensorflow-gpu
-      ci_event: Daily CI
-    secrets: inherit
+  # tf-pipeline:
+  #   name: TF pipeline CI
+  #   uses: ./.github/workflows/self-scheduled.yml
+  #   with:
+  #     job: run_pipelines_tf_gpu
+  #     slack_report_channel: "#transformers-ci-daily-pipeline-tf"
+  #     runner: daily-ci
+  #     docker: huggingface/transformers-tensorflow-gpu
+  #     ci_event: Daily CI
+  #   secrets: inherit
 
-  example-ci:
-    name: Example CI
-    uses: ./.github/workflows/self-scheduled.yml
-    with:
-      job: run_examples_gpu
-      slack_report_channel: "#transformers-ci-daily-examples"
-      runner: daily-ci
-      docker: huggingface/transformers-all-latest-gpu
-      ci_event: Daily CI
-    secrets: inherit
+  # example-ci:
+  #   name: Example CI
+  #   uses: ./.github/workflows/self-scheduled.yml
+  #   with:
+  #     job: run_examples_gpu
+  #     slack_report_channel: "#transformers-ci-daily-examples"
+  #     runner: daily-ci
+  #     docker: huggingface/transformers-all-latest-gpu
+  #     ci_event: Daily CI
+  #   secrets: inherit
 
-  deepspeed-ci:
-    name: DeepSpeed CI
-    uses: ./.github/workflows/self-scheduled.yml
-    with:
-      job: run_torch_cuda_extensions_gpu
-      slack_report_channel: "#transformers-ci-daily-deepspeed"
-      runner: daily-ci
-      docker: huggingface/transformers-pytorch-deepspeed-latest-gpu
-      ci_event: Daily CI
-      working-directory-prefix: /workspace
-    secrets: inherit
+  # deepspeed-ci:
+  #   name: DeepSpeed CI
+  #   uses: ./.github/workflows/self-scheduled.yml
+  #   with:
+  #     job: run_torch_cuda_extensions_gpu
+  #     slack_report_channel: "#transformers-ci-daily-deepspeed"
+  #     runner: daily-ci
+  #     docker: huggingface/transformers-pytorch-deepspeed-latest-gpu
+  #     ci_event: Daily CI
+  #     working-directory-prefix: /workspace
+  #   secrets: inherit
 
   quantization-ci:
     name: Quantization CI

src/transformers/utils/import_utils.py

@@ -1002,8 +1002,18 @@ def is_compressed_tensors_available():
 
 
 def is_auto_gptq_available():
-    return _auto_gptq_available
+    if not _auto_gptq_available:
+        return _auto_gptq_available
 
+    try:
+        from auto_gptq import exllama_set_max_input_length
+    except ImportError as exc:
+        if "shard_checkpoint" in str(exc):
+            # auto_gptq requires eetq and it is currently broken with newer transformers versions because it tries to import shard_checkpoint
+            # see https://github.com/NetEase-FuXi/EETQ/issues/34
+            # TODO: Remove once eetq releasees a fix and this release is used in CI
+            return False
+    return _auto_gptq_available
 
 def is_eetq_available():
     return _eetq_available