[CI] [GHA] Use snapshot_download to load HF models

.github/workflows/job_jax_layer_tests.yml

@@ -48,6 +48,8 @@ jobs:
       INSTALL_DIR: ${{ github.workspace }}/install
       INSTALL_TEST_DIR: ${{ github.workspace }}/install/tests
       USE_SYSTEM_CACHE: False # Using remote HuggingFace cache
+      HF_HUB_VERBOSITY: debug
+      TRANSFORMERS_VERBOSITY: debug
     steps:
       # checkout action cleans up the workspace and have to be the first step
       - name: Fetch custom actions
@@ -75,7 +77,6 @@ jobs:
         run: |
           echo "HF_HUB_CACHE=${{ runner.os == 'Linux' && env.HF_HUB_CACHE_LIN || env.HF_HUB_CACHE_WIN }}" >> "$GITHUB_ENV"
           echo "HUGGINGFACE_HUB_CACHE=${{ runner.os == 'Linux' && env.HF_HUB_CACHE_LIN || env.HF_HUB_CACHE_WIN }}" >> "$GITHUB_ENV"
-          echo "HF_HUB_OFFLINE=1" >> "$GITHUB_ENV"
 
       - name: Install OpenVINO dependencies (mac)
         if: runner.os == 'macOS'

.github/workflows/job_jax_models_tests.yml

@@ -44,6 +44,8 @@ jobs:
       INSTALL_TEST_DIR: ${{ github.workspace }}/install/tests
       MODEL_HUB_TESTS_INSTALL_DIR: ${{ github.workspace }}/install/tests/model_hub_tests
       USE_SYSTEM_CACHE: False # Using remote HuggingFace cache
+      HF_HUB_VERBOSITY: debug
+      TRANSFORMERS_VERBOSITY: debug
     steps:
       # checkout action cleans up the workspace and have to be the first step
       - name: Fetch custom actions
@@ -77,7 +79,6 @@ jobs:
         run: |
           echo "HF_HUB_CACHE=${{ runner.os == 'Linux' && env.HF_HUB_CACHE_LIN || env.HF_HUB_CACHE_WIN }}" >> "$GITHUB_ENV"
           echo "HUGGINGFACE_HUB_CACHE=${{ runner.os == 'Linux' && env.HF_HUB_CACHE_LIN || env.HF_HUB_CACHE_WIN }}" >> "$GITHUB_ENV"
-          echo "HF_HUB_OFFLINE=1" >> "$GITHUB_ENV"
 
       - name: Extract OpenVINO packages and tests
         run: pigz -dc openvino_tests.tar.gz | tar -xf - -v

.github/workflows/job_onnx_models_tests.yml

@@ -46,6 +46,8 @@ jobs:
       ONNX_MODEL_ZOO_SHA: "5faef4c33eba0395177850e1e31c4a6a9e634c82"
       USE_SYSTEM_CACHE: False # Using remote HuggingFace cache
       LOGS_FOLDER: ${{ github.workspace }}/onnx_models_tests_logs"
+      HF_HUB_VERBOSITY: debug
+      TRANSFORMERS_VERBOSITY: debug
     if: ${{ github.event_name != 'merge_group' }}
     steps:
       # checkout action cleans up the workspace and have to be the first step
@@ -78,7 +80,6 @@ jobs:
         run: |
           echo "HF_HUB_CACHE=${{ runner.os == 'Linux' && env.HF_HUB_CACHE_LIN || env.HF_HUB_CACHE_WIN }}" >> "$GITHUB_ENV"
           echo "HUGGINGFACE_HUB_CACHE=$HF_HUB_CACHE" >> "$GITHUB_ENV"
-          echo "HF_HUB_OFFLINE=1" >> "$GITHUB_ENV"
           echo "MODELS_SHARE_PATH=/mount/testdata$((GITHUB_RUN_NUMBER % NUMBER_OF_REPLICAS))" >> "$GITHUB_ENV"
 
       # Issue 148922

.github/workflows/job_pytorch_fx_layer_tests.yml

@@ -47,6 +47,8 @@ jobs:
       INSTALL_DIR: ${{ github.workspace }}/install
       INSTALL_TEST_DIR: ${{ github.workspace }}/install/tests
       USE_SYSTEM_CACHE: False # Using remote HuggingFace cache
+      HF_HUB_VERBOSITY: debug
+      TRANSFORMERS_VERBOSITY: debug
     steps:
       - name: Fetch setup_python and install wheels actions
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
@@ -74,7 +76,6 @@ jobs:
         run: |
           Add-Content -Path $env:GITHUB_ENV -Value "HF_HUB_CACHE=C:\\mount\\caches\\huggingface"
           Add-Content -Path $env:GITHUB_ENV -Value "HUGGINGFACE_HUB_CACHE=C:\\mount\\caches\\huggingface"
-          Add-Content -Path $env:GITHUB_ENV -Value "HF_HUB_OFFLINE=1"
 
       - name: Install OpenVINO dependencies (mac)
         if: runner.os == 'macOS'

.github/workflows/job_pytorch_layer_tests.yml

@@ -49,6 +49,8 @@ jobs:
       INSTALL_TEST_DIR: ${{ github.workspace }}/install/tests
       LAYER_TESTS_INSTALL_DIR: ${{ github.workspace }}/install/tests/layer_tests
       USE_SYSTEM_CACHE: False # Using remote HuggingFace cache
+      HF_HUB_VERBOSITY: debug
+      TRANSFORMERS_VERBOSITY: debug
     steps:
       - name: Fetch setup_python and install wheels actions
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
@@ -75,7 +77,7 @@ jobs:
         run: |
           echo "HF_HUB_CACHE=${{ runner.os == 'Linux' && env.HF_HUB_CACHE_LIN || env.HF_HUB_CACHE_WIN }}" >> "$GITHUB_ENV"
           echo "HUGGINGFACE_HUB_CACHE=${{ runner.os == 'Linux' && env.HF_HUB_CACHE_LIN || env.HF_HUB_CACHE_WIN }}" >> "$GITHUB_ENV"
-          echo "HF_HUB_OFFLINE=1" >> "$GITHUB_ENV"
+
       - name: Install OpenVINO dependencies (mac)
         if: runner.os == 'macOS'
         run: brew install pigz

.github/workflows/job_pytorch_models_tests.yml

@@ -43,6 +43,8 @@ jobs:
       INSTALL_TEST_DIR: ${{ github.workspace }}/install/tests
       MODEL_HUB_TESTS_INSTALL_DIR: ${{ github.workspace }}/install/tests/model_hub_tests
       USE_SYSTEM_CACHE: False # Using remote HuggingFace cache
+      HF_HUB_VERBOSITY: debug
+      TRANSFORMERS_VERBOSITY: debug
     steps:
       - name: Fetch setup_python and install wheels actions
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
@@ -75,7 +77,6 @@ jobs:
         run: |
           echo "HF_HUB_CACHE=${{ runner.os == 'Linux' && env.HF_HUB_CACHE_LIN || env.HF_HUB_CACHE_WIN }}" >> "$GITHUB_ENV"
           echo "HUGGINGFACE_HUB_CACHE=${{ runner.os == 'Linux' && env.HF_HUB_CACHE_LIN || env.HF_HUB_CACHE_WIN }}" >> "$GITHUB_ENV"
-          echo "HF_HUB_OFFLINE=1" >> "$GITHUB_ENV"
 
       - name: Extract OpenVINO artifacts
         run: pigz -dc openvino_tests.tar.gz | tar -xf - -v
@@ -119,7 +120,6 @@ jobs:
           TYPE: ${{ inputs.model_scope == 'precommit' && 'precommit' || 'nightly' }}
           TEST_DEVICE: CPU
           OP_REPORT_FILE: ${{ env.INSTALL_TEST_DIR }}/TEST-torch_unsupported_ops.log
-          HF_HUB_OFFLINE: 0  # Can't use offline mode - these tests use `model_info` call, which can't be cached
 
       - name: PagedAttention Test
         if: ${{ inputs.model_scope == 'precommit' }}
@@ -128,7 +128,6 @@ jobs:
           python3 -m pytest ${MODEL_HUB_TESTS_INSTALL_DIR}/transformation_tests/test_pa_transformation.py -m precommit --html=${INSTALL_TEST_DIR}/TEST-torch_pagedattention_tests.html --self-contained-html -vvv -s --tb=short -n 2
         env:
           TEST_DEVICE: CPU
-          HF_HUB_OFFLINE: 0
 
       - name: RoPE Test
         if: ${{ inputs.model_scope == 'precommit' }}

.github/workflows/job_tensorflow_layer_tests.yml

@@ -48,6 +48,8 @@ jobs:
       INSTALL_DIR: ${{ github.workspace }}/install
       INSTALL_TEST_DIR: ${{ github.workspace }}/install/tests
       USE_SYSTEM_CACHE: False # Using remote HuggingFace cache
+      HF_HUB_VERBOSITY: debug
+      TRANSFORMERS_VERBOSITY: debug
     steps:
       - name: Fetch custom actions
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
@@ -80,7 +82,6 @@ jobs:
         run: |
           echo "HF_HUB_CACHE=${{ runner.os == 'Linux' && env.HF_HUB_CACHE_LIN || env.HF_HUB_CACHE_WIN }}" >> "$GITHUB_ENV"
           echo "HUGGINGFACE_HUB_CACHE=${{ runner.os == 'Linux' && env.HF_HUB_CACHE_LIN || env.HF_HUB_CACHE_WIN }}" >> "$GITHUB_ENV"
-          echo "HF_HUB_OFFLINE=1" >> "$GITHUB_ENV"
 
       - name: Install OpenVINO dependencies (mac)
         if: ${{ runner.os == 'macOS' }}

.github/workflows/job_tensorflow_models_tests.yml

@@ -43,6 +43,8 @@ jobs:
       INSTALL_TEST_DIR: ${{ github.workspace }}/install/tests
       NUMBER_OF_REPLICAS: 2
       USE_SYSTEM_CACHE: False # Using remote HuggingFace cache
+      HF_HUB_VERBOSITY: debug
+      TRANSFORMERS_VERBOSITY: debug
     steps:
       - name: Fetch custom actions
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
@@ -76,7 +78,6 @@ jobs:
           echo "HF_HUB_CACHE=${{ runner.os == 'Linux' && env.HF_HUB_CACHE_LIN || env.HF_HUB_CACHE_WIN }}" >> "$GITHUB_ENV"
           echo "HUGGINGFACE_HUB_CACHE=${{ runner.os == 'Linux' && env.HF_HUB_CACHE_LIN || env.HF_HUB_CACHE_WIN }}" >> "$GITHUB_ENV"
           echo "TFHUB_CACHE_DIR=/mount/testdata$((GITHUB_RUN_NUMBER % NUMBER_OF_REPLICAS))/tfhub_models" >> "$GITHUB_ENV"
-          echo "HF_HUB_OFFLINE=1" >> "$GITHUB_ENV"
 
       - name: Extract OpenVINO artifacts (Linux and macOS)
         run: pigz -dc openvino_tests.tar.gz | tar -xf - -v

tests/e2e_tests/test_utils/pytorch_loaders.py

@@ -6,6 +6,7 @@
 import sys
 
 import torch
+from huggingface_hub import snapshot_download
 
 
 class LoadPyTorchModel:
@@ -70,7 +71,8 @@ def load_cadene_model(module, args):
 
 def load_hugging_face_model(module, args):
     module = importlib.import_module(module)
-    model = module.AutoModel.from_pretrained(args['model-name'], torchscript=True)
+    model_cached = snapshot_download(args['model-name'])  # required to avoid HF rate limits
+    model = module.AutoModel.from_pretrained(model_cached, torchscript=True)
 
     return model
 

tests/llm/accuracy_conformance.py

@@ -5,6 +5,7 @@
 import tempfile
 
 import pytest
+from huggingface_hub import snapshot_download
 from optimum.intel.openvino import (OVModelForCausalLM,
                                     OVWeightQuantizationConfig)
 from transformers import AutoModelForCausalLM, AutoTokenizer, set_seed
@@ -100,8 +101,9 @@ def setup_model(model_id):
     logger.info(f"Setting up model: {model_id}")
 
     # Download original model
-    model = AutoModelForCausalLM.from_pretrained(model_id)
-    tokenizer = AutoTokenizer.from_pretrained(model_id)
+    model_cached = snapshot_download(model_id)  # required to avoid HF rate limits
+    model = AutoModelForCausalLM.from_pretrained(model_cached)
+    tokenizer = AutoTokenizer.from_pretrained(model_cached)
 
     # Save original model
     model_path = get_model_path(model_id, "org")

tests/model_hub_tests/jax/test_hf_transformers.py

@@ -5,6 +5,7 @@
 import os
 import pytest
 import requests
+from huggingface_hub import snapshot_download
 from PIL import Image
 from models_hub_common.constants import hf_cache_dir, clean_hf_cache_dir
 from models_hub_common.utils import cleanup_dir, get_models_list, retry
@@ -22,17 +23,18 @@
 class TestTransformersModel(TestJaxConvertModel):
     @retry(3, exceptions=(OSError,), delay=1)
     def load_model(self, model_name, _):
-        model = FlaxAutoModel.from_pretrained(model_name)
+        model_cached = snapshot_download(model_name)  # required to avoid HF rate limits
+        model = FlaxAutoModel.from_pretrained(model_cached)
         if model_name in ['google/vit-base-patch16-224-in21k']:
             url = "http://images.cocodataset.org/val2017/000000039769.jpg"
             image = Image.open(requests.get(url, stream=True).raw)
-            image_processor = AutoImageProcessor.from_pretrained(model_name)
+            image_processor = AutoImageProcessor.from_pretrained(model_cached)
             self.example = image_processor(images=image, return_tensors="np")
         elif model_name in ['albert/albert-base-v2', 'facebook/bart-base', 'ksmcg/Mistral-tiny']:
-            tokenizer = AutoTokenizer.from_pretrained(model_name)
+            tokenizer = AutoTokenizer.from_pretrained(model_cached)
             self.example = tokenizer("Hello, my dog is cute", return_tensors="np")
         elif model_name in ['openai/clip-vit-base-patch32']:
-            processor = AutoProcessor.from_pretrained(model_name)
+            processor = AutoProcessor.from_pretrained(model_cached)
             url = "http://images.cocodataset.org/val2017/000000039769.jpg"
             image = Image.open(requests.get(url, stream=True).raw)
             self.example = processor(text=["a photo of a cat", "a photo of a dog"],

tests/model_hub_tests/pytorch/test_edsr.py

@@ -6,6 +6,7 @@
 import pytest
 import random
 import torch
+from huggingface_hub import snapshot_download
 from models_hub_common.constants import hf_cache_dir, clean_hf_cache_dir
 from models_hub_common.utils import cleanup_dir
 
@@ -44,8 +45,9 @@ def load_model(self, model_name, model_link):
         image = Image.open(requests.get(url, stream=True).raw)
         assert model_name in name_to_class, "Unexpected model name"
         print(f"scale: {self.scale}")
+        model_cached = snapshot_download(f'eugenesiow/{model_name}')  # required to avoid HF rate limits
         model = name_to_class[model_name].from_pretrained(
-            f'eugenesiow/{model_name}', scale=self.scale)
+            model_cached, scale=self.scale)
         inputs = ImageLoader.load_image(image)
         self.example = (torch.randn_like(inputs),)
         self.inputs = (inputs,)