[misc] lint (#9593)

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

src/llamafactory/model/model_utils/checkpointing.py

@@ -18,8 +18,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import os
 import inspect
+import os
 from functools import WRAPPER_ASSIGNMENTS, partial, wraps
 from types import MethodType
 from typing import TYPE_CHECKING, Any, Callable, Optional, Union
@@ -158,9 +158,7 @@ def prepare_model_for_training(model: "PreTrainedModel", model_args: "ModelArgum
         and int(os.environ.get("FSDP_VERSION", "1")) == 2
     ):
         model_args.use_reentrant_gc = False
-        logger.warning_rank0(
+        logger.warning_rank0("You are using fsdp2, `use_reentrant_gc` has been set to False.")
-            "You are using fsdp2, `use_reentrant_gc` has been set to False. "
-        )
 
     if not model_args.disable_gradient_checkpointing:
         if not getattr(model, "supports_gradient_checkpointing", False):

src/llamafactory/train/sft/workflow.py

@@ -28,6 +28,7 @@ from ..trainer_utils import create_modelcard_and_push
 from .metric import ComputeAccuracy, ComputeSimilarity, eval_logit_processor
 from .trainer import CustomSeq2SeqTrainer
 
+
 if TYPE_CHECKING:
     from transformers import Seq2SeqTrainingArguments, TrainerCallback
 

src/llamafactory/v1/extras/__init__.py

@@ -1 +0,0 @@
-

tests/conftest.py

@@ -40,9 +40,7 @@ def pytest_configure(config):
     config.addinivalue_line(
         "markers", "require_device: test requires specific device, e.g., @pytest.mark.require_device('cuda')"
     )
-    config.addinivalue_line(
+    config.addinivalue_line("markers", "runs_on: test requires specific device, e.g., @pytest.mark.runs_on(['cpu'])")
-        "markers", "runs_on: test requires specific device, e.g., @pytest.mark.runs_on(['cpu'])"
-    )
 
 
 def _handle_runs_on(items):
@@ -64,14 +62,12 @@ def _handle_runs_on(items):
             if isinstance(runs_on_devices, str):
                 runs_on_devices = [runs_on_devices]
 
-
             if CURRENT_DEVICE not in runs_on_devices:
                 item.add_marker(
-                    pytest.mark.skip(
+                    pytest.mark.skip(reason=f"test requires one of {runs_on_devices} (current: {CURRENT_DEVICE})")
-                        reason=f"test requires one of {runs_on_devices} (current: {CURRENT_DEVICE})"
-                    )
                 )
 
+
 def _handle_slow_tests(items):
     """Skip slow tests unless RUN_SLOW environment variable is set.
 

tests/data/test_template.py

@@ -284,7 +284,6 @@ def test_llama4_template(use_fast: bool):
         pytest.param(False, marks=pytest.mark.xfail(reason="Phi-4 slow tokenizer is broken.")),
     ],
 )
-
 @pytest.mark.runs_on(["cpu"])
 def test_phi4_template(use_fast: bool):
     prompt_str = (

tests/e2e/test_train.py

@@ -48,6 +48,7 @@ INFER_ARGS = {
 
 OS_NAME = os.getenv("OS_NAME", "")
 
+
 @pytest.mark.runs_on(["cpu"])
 @pytest.mark.parametrize(
     "stage,dataset",

tests/eval/test_eval_template.py

@@ -55,6 +55,7 @@ def test_eval_template_en():
         {"role": "assistant", "content": "C"},
     ]
 
+
 @pytest.mark.runs_on(["cpu"])
 def test_eval_template_zh():
     support_set = [

tests/model/model_utils/test_attention.py

@@ -17,13 +17,16 @@ import os
 import pytest
 from transformers.utils import is_flash_attn_2_available
 
+
 # Compatible with Transformers v4 and Transformers v5
 try:
     from transformers.utils import is_torch_sdpa_available
 except ImportError:
+
     def is_torch_sdpa_available():
         return True
 
+
 from llamafactory.extras.packages import is_transformers_version_greater_than
 from llamafactory.train.test_utils import load_infer_model
 

tests/model/test_base.py

@@ -29,6 +29,7 @@ INFER_ARGS = {
     "infer_dtype": "float16",
 }
 
+
 @pytest.mark.runs_on(["cpu", "npu"])
 @pytest.mark.skip_on_devices("npu")
 def test_base():
@@ -36,6 +37,7 @@ def test_base():
     ref_model = load_reference_model(TINY_LLAMA3)
     compare_model(model, ref_model)
 
+
 @pytest.mark.runs_on(["cpu", "npu"])
 @pytest.mark.skip_on_devices("npu")
 @pytest.mark.usefixtures("fix_valuehead_cpu_loading")

tests/model/test_full.py

@@ -43,6 +43,7 @@ INFER_ARGS = {
     "infer_dtype": "float16",
 }
 
+
 @pytest.mark.runs_on(["cpu", "npu"])
 def test_full_train():
     model = load_train_model(**TRAIN_ARGS)
@@ -50,6 +51,7 @@ def test_full_train():
         assert param.requires_grad is True
         assert param.dtype == torch.float32
 
+
 @pytest.mark.runs_on(["cpu", "npu"])
 def test_full_inference():
     model = load_infer_model(**INFER_ARGS)

tests/model/test_lora.py

@@ -102,6 +102,7 @@ def test_lora_train_valuehead():
     assert torch.allclose(state_dict["v_head.summary.weight"], ref_state_dict["v_head.summary.weight"])
     assert torch.allclose(state_dict["v_head.summary.bias"], ref_state_dict["v_head.summary.bias"])
 
+
 @pytest.mark.runs_on(["cpu", "npu"])
 @pytest.mark.skip_on_devices("npu")
 def test_lora_inference():

tests/model/test_pissa.py

@@ -56,6 +56,7 @@ def test_pissa_train():
     ref_model = load_reference_model(TINY_LLAMA_PISSA, TINY_LLAMA_PISSA, use_pissa=True, is_trainable=True)
     compare_model(model, ref_model)
 
+
 @pytest.mark.runs_on(["cpu", "npu"])
 @pytest.mark.xfail(reason="Known connection error.")
 def test_pissa_inference():