Improve Hugging Face API utilization in tests (#473)

* Remove isolation of downloaded model

* Use k8s-util runners with model cache and API key

* Minor cleanup in 'test' action

Author: dbarbuzzi

.github/actions/test/action.yml

@@ -9,8 +9,7 @@ inputs:
     required: true
   code_coverage:
     description: whether to collect code coverage metrics during test run
-    type: boolean
-    default: false
+    default: 'false'
 outputs:
   status:
     description: "final status from test"
@@ -51,7 +50,7 @@ runs:
 
           if [[ "${ENABLE_COVERAGE}" == "true" ]]; then
             echo "::group::Installing code coverage requirements via pip"
-            pip install bashlex https://github.com/neuralmagic/pytest-nm-releng/archive/v0.4.0.tar.gz
+            pip install https://github.com/neuralmagic/pytest-nm-releng/archive/v0.4.0.tar.gz
             pip install coverage pytest-cov
 
             # Adding Code coverage to the tests
@@ -76,7 +75,7 @@ runs:
             fi
             echo "::endgroup::"
           fi
-          
+
           deactivate
           exit ${SUCCESS}
       shell: bash

.github/workflows/test-check.yaml

@@ -12,7 +12,10 @@ on:
 
 jobs:
   python-tests:
-    runs-on: ubuntu-24.04
+    runs-on: k8s-util
+    env:
+        HF_HOME: /model-cache
+        HF_TOKEN: ${{ secrets.NM_HF_TOKEN_READ_ONLY }}
     steps:
         - uses: actions/setup-python@v5
           with:
@@ -21,11 +24,13 @@ jobs:
           with:
             fetch-depth: 0
             fetch-tags: true
+        - name: Install system dependencies
+          run: |-
+            sudo apt-get update
+            sudo apt-get install -y --no-install-recommends g++ gcc make
         - name: Set Env
-          run: |
-            pip3 install --upgrade pip && pip3 install --upgrade setuptools
+          run: pip3 install --upgrade pip setuptools
         - name: "⚙️ Install dependencies"
           run: pip3 install .[dev,accelerate]
         - name: "🔬 Running tests"
           run: make test
-

.github/workflows/test.yml

@@ -71,6 +71,9 @@ jobs:
             contents: 'read'
             id-token: 'write'
             pages: 'write'
+        env:
+            HF_HOME: /model-cache
+            HF_TOKEN: ${{ secrets.NM_HF_TOKEN_READ_ONLY }}
         environment:
             name: github-pages
             url: ${{ steps.coverage.outputs.page_url }}
@@ -90,13 +93,9 @@ jobs:
                   python-version: ${{ inputs.python }}
 
             - name: install system dependencies
-              run: |
-                if command -v g++ >/dev/null 2>&1; then
-                  echo "found g++ compiler"
-                else
-                   echo "installing g++ etc compilers..."
-                   sudo apt update && sudo apt install -y g++ gcc
-                fi
+              run: |-
+                  sudo apt-get update
+                  sudo apt-get install -y --no-install-recommends g++ gcc make
               shell: bash
 
             - name: checkout code

.github/workflows/trigger-all.yml

@@ -19,7 +19,7 @@ on:
                 type: boolean
                 default: false
             gitref:
-                description: "git commit hash or tag name" 
+                description: "git commit hash or tag name"
                 type: string
                 default: 'main'
 
@@ -32,8 +32,8 @@ jobs:
             wf_category: ${{ inputs.wf_category || 'NIGHTLY' }}
             gitref: ${{ inputs.gitref || 'main' }}
             push_to_pypi: ${{ (github.event.schedule == '30 0 * * *') || inputs.push_to_pypi || false }}
-            test_configs: '[{"python":"3.11.4","label":"ubuntu-24.04","timeout":"40","code_coverage":true},
-                            {"python":"3.10.12","label":"ubuntu-22.04","timeout":"40"},
+            test_configs: '[{"python":"3.11.4","label":"k8s-util","timeout":"40","code_coverage":true},
+                            {"python":"3.10.12","label":"k8s-util","timeout":"40"},
                             {"python":"3.9.17","label":"k8s-h100-solo","timeout":"40"},
                             {"python":"3.12.6","label":"k8s-a100-duo","timeout":"40"}]'
 

tests/test_utils/test_helpers.py

@@ -36,12 +36,11 @@ def tensors():
 
 
 @pytest.fixture
-def llama_model(tmp_path):
-    model_name = "neuralmagic/llama2.c-stories110M-pruned50"
-    model = AutoModelForCausalLM.from_pretrained(
-        model_name, torch_dtype="auto", cache_dir=tmp_path
+def llama_model():
+    return AutoModelForCausalLM.from_pretrained(
+        "RedHatAI/llama2.c-stories110M-pruned50",
+        torch_dtype="auto",
     )
-    yield model
 
 
 def test_save_compressed_sparse_bitmask(tmp_path, tensors):
@@ -120,9 +119,9 @@ def test_load_compressed_dense(tmp_path, tensors):
 
 
 def test_load_compressed_sharded(tmp_path, llama_model):
-    sharded_model_path = tmp_path / "shared_model"
+    sharded_model_path = tmp_path / "sharded_model"
     llama_model.save_pretrained(sharded_model_path, max_shard_size="2MB")
-    # make sure that model is shared on disk
+    # make sure that model is sharded on disk
     assert len(os.listdir(sharded_model_path)) > 1
     loaded_state_dict = dict(load_compressed(sharded_model_path))
     for key, value in llama_model.state_dict().items():