[v1] model loader (#9613)

tests_v1/accelerator/test_interface.py

@@ -15,11 +15,11 @@
 
 import os
 
-from llamafactory.v1.accelerator.interface import DistributedInterface, DistributedStrategy
+from llamafactory.v1.accelerator.interface import DistributedInterface
 
 
 def test_distributed_interface():
-    DistributedInterface(DistributedStrategy())
+    DistributedInterface()
     assert DistributedInterface.get_rank() == int(os.getenv("RANK", "0"))
     assert DistributedInterface.get_world_size() == int(os.getenv("WORLD_SIZE", "1"))
     assert DistributedInterface.get_local_rank() == int(os.getenv("LOCAL_RANK", "0"))

tests_v1/conftest.py

@@ -0,0 +1,29 @@
+# Copyright 2025 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+from pytest import Config, Item
+
+from llamafactory.v1.utils.packages import is_transformers_version_greater_than
+
+
+def pytest_collection_modifyitems(config: Config, items: list[Item]):
+    if is_transformers_version_greater_than("4.57.0"):
+        return
+
+    skip_bc = pytest.mark.skip(reason="Skip backward compatibility tests")
+
+    for item in items:
+        if "tests_v1" in str(item.fspath):
+            item.add_marker(skip_bc)

tests_v1/core/test_model_loader.py

@@ -0,0 +1,33 @@
+# Copyright 2025 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import torch
+
+from llamafactory.v1.config.model_args import ModelArguments
+from llamafactory.v1.core.model_loader import ModelLoader
+
+
+def test_tiny_qwen():
+    from transformers import Qwen2Config, Qwen2ForCausalLM, Qwen2TokenizerFast
+
+    model_args = ModelArguments(model="llamafactory/tiny-random-qwen2.5")
+    model_loader = ModelLoader(model_args)
+    assert isinstance(model_loader.processor, Qwen2TokenizerFast)
+    assert isinstance(model_loader.model.config, Qwen2Config)
+    assert isinstance(model_loader.model, Qwen2ForCausalLM)
+    assert model_loader.model.dtype == torch.bfloat16
+
+
+if __name__ == "__main__":
+    test_tiny_qwen()

tests_v1/plugins/data_plugins/test_converter.py

@@ -24,7 +24,7 @@ from llamafactory.v1.plugins.data_plugins.converter import DataConverterPlugin
 
 @pytest.mark.parametrize("num_samples", [16])
 def test_alpaca_converter(num_samples: int):
-    data_args = DataArguments(dataset="llamafactory/v1-sft-demo/dataset_info.yaml")
+    data_args = DataArguments(dataset="llamafactory/v1-dataset-info/tiny-supervised-dataset.yaml")
     data_engine = DataEngine(data_args)
     original_data = load_dataset("llamafactory/tiny-supervised-dataset", split="train")
     indexes = random.choices(range(len(data_engine)), k=num_samples)
@@ -54,6 +54,8 @@ def test_sharegpt_converter():
         "conversations": [
             {"from": "system", "value": "System"},
             {"from": "human", "value": "User"},
+            {"from": "function_call", "value": "Tool"},
+            {"from": "observation", "value": "Observation"},
             {"from": "gpt", "value": "Assistant"},
         ]
     }
@@ -61,6 +63,8 @@ def test_sharegpt_converter():
         "messages": [
             {"content": [{"type": "text", "value": "System"}], "loss_weight": 0.0, "role": "system"},
             {"content": [{"type": "text", "value": "User"}], "loss_weight": 0.0, "role": "user"},
+            {"content": [{"type": "tool_calls", "value": "Tool"}], "loss_weight": 1.0, "role": "assistant"},
+            {"content": [{"type": "text", "value": "Observation"}], "loss_weight": 0.0, "role": "tool"},
             {"content": [{"type": "text", "value": "Assistant"}], "loss_weight": 1.0, "role": "assistant"},
         ]
     }
@@ -69,7 +73,7 @@ def test_sharegpt_converter():
 
 @pytest.mark.parametrize("num_samples", [16])
 def test_pair_converter(num_samples: int):
-    data_args = DataArguments(dataset="llamafactory/tiny-preference-dataset/dataset_info.yaml")
+    data_args = DataArguments(dataset="llamafactory/v1-dataset-info/orca-dpo-pairs.yaml")
     data_engine = DataEngine(data_args)
     original_data = load_dataset("HuggingFaceH4/orca_dpo_pairs", split="train_prefs")
     indexes = random.choices(range(len(data_engine)), k=num_samples)
@@ -112,7 +116,7 @@ def test_pair_converter(num_samples: int):
                 },
             ],
         }
-        assert data_engine[index] == {"_dataset_name": "dpo_zh_demo", **expected_data}
+        assert data_engine[index] == {"_dataset_name": "tiny_dataset", **expected_data}
 
 
 if __name__ == "__main__":