ci: try to fix test-full (#868)

* ci: try to fix test-full

also use 3.12 to run full tests

* fix mypy errors

* hf: fix deprecated arguments in transformers.TrainingArguments

* tests: adjust assertions in test_huggingface_log_model

Due to huggingface/transformers#36219.

Author: skshetry

.github/workflows/tests.yml

@@ -27,32 +27,27 @@ jobs:
       - name: Set up Python
         uses: actions/setup-python@v5
         with:
-          python-version: "3.9"
-          cache: "pip"
-          cache-dependency-path: setup.cfg
+          python-version: "3.12"
 
-      - name: Upgrade pip
-        run: |
-          python -m pip install --upgrade pip wheel
-          pip --version
+      - uses: astral-sh/setup-uv@v6
+        with:
+          enable-cache: true
+          cache-suffix: ${{ matrix.pyv }}
+          cache-dependency-glob: pyproject.toml
 
-      - run: python -m pip install pre-commit
+      - name: Full install
+        run: uv pip install -e '.[dev]' --system
 
       - uses: actions/cache@v4
         with:
           path: ~/.cache/pre-commit/
           key: pre-commit-4|${{ env.pythonLocation }}|${{ hashFiles('.pre-commit-config.yaml') }}
 
       - name: pre-commit
-        run: |
-          pre-commit run --show-diff-on-failure --color=always --all-files
-
-      - name: Full install
-        run: pip install -e '.[dev]'
+        run: uv tool run pre-commit run --show-diff-on-failure --color=always --all-files
 
       - name: mypy
-        run: |
-          python -m mypy
+        run: mypy
 
       - name: Run tests
         run: pytest -v tests --cov --cov-report=xml --cov-config=pyproject.toml

examples/DVCLive-HuggingFace.ipynb

@@ -200,7 +200,7 @@
     "lr = 3e-4\n",
     "\n",
     "training_args = TrainingArguments(\n",
-    "    evaluation_strategy=\"epoch\",\n",
+    "    eval_strategy=\"epoch\",\n",
     "    learning_rate=lr,\n",
     "    logging_strategy=\"epoch\",\n",
     "    num_train_epochs=5,\n",
@@ -243,7 +243,7 @@
     "lr = 1e-4\n",
     "\n",
     "training_args = TrainingArguments(\n",
-    "    evaluation_strategy=\"epoch\",\n",
+    "    eval_strategy=\"epoch\",\n",
     "    learning_rate=lr,\n",
     "    logging_strategy=\"epoch\",\n",
     "    num_train_epochs=5,\n",

src/dvclive/fabric.py

@@ -90,8 +90,8 @@ def log_metrics(
 
         for metric_name, metric_val in metrics.items():
             val = metric_val
-            if is_tensor(val):
-                val = val.cpu().detach().item()  # type: ignore[union-attr]
+            if is_tensor(val):  # type: ignore[unreachable]
+                val = val.cpu().detach().item()  # type: ignore[union-attr,unreachable]
             name = standardize_metric_name(metric_name, __name__)
             if Metric.could_log(val):
                 self.experiment.log_metric(name=name, val=val)

src/dvclive/huggingface.py

@@ -62,6 +62,7 @@ def on_save(
         **kwargs,
     ):
         if self._log_model == "all" and state.is_world_process_zero:
+            assert args.output_dir is not None  # noqa: S101
             self.live.log_artifact(args.output_dir)
 
     def on_train_end(
@@ -79,6 +80,7 @@ def on_train_end(
                 eval_dataset=["fake"],
             )
             name = "best" if args.load_best_model_at_end else "last"
+            assert args.output_dir is not None  # noqa: S101
             output_dir = os.path.join(args.output_dir, name)
             fake_trainer.save_model(output_dir)
             self.live.log_artifact(output_dir, name=name, type="model", copy=True)

tests/frameworks/test_huggingface.py

@@ -99,7 +99,7 @@ def model():
 def args():
     return TrainingArguments(
         "foo",
-        evaluation_strategy="epoch",
+        eval_strategy="epoch",
         num_train_epochs=2,
         save_strategy="epoch",
         report_to="none",  # Disable auto-reporting to avoid duplication
@@ -175,7 +175,7 @@ def test_huggingface_log_model(
     trainer.train()
 
     expected_call_count = {
-        "all": 3,
+        "all": 2,
         True: 1,
         False: 0,
         None: 0,