Fix chat template not applied in TransformersLLM (argilla-io#1083)

.github/workflows/docs-pr-close.yml

@@ -8,6 +8,10 @@ concurrency:
   group: distilabel-docs
   cancel-in-progress: false
 
+permissions:
+  contents: write
+  pull-requests: write
+
 jobs:
   cleanup:
     runs-on: ubuntu-latest

.github/workflows/docs-pr.yml

@@ -10,6 +10,10 @@ concurrency:
   group: distilabel-docs
   cancel-in-progress: false
 
+permissions:
+  contents: write
+  pull-requests: write
+
 jobs:
   publish:
     runs-on: ubuntu-latest

.github/workflows/docs.yml

@@ -12,6 +12,10 @@ concurrency:
   group: distilabel-docs
   cancel-in-progress: false
 
+permissions:
+  contents: write
+  pull-requests: write
+
 jobs:
   publish:
     runs-on: ubuntu-latest

src/distilabel/__init__.py

@@ -14,6 +14,6 @@
 
 from rich import traceback as rich_traceback
 
-__version__ = "1.4.1"
+__version__ = "1.4.2"
 
 rich_traceback.install(show_locals=True)

src/distilabel/llms/huggingface/transformers.py

@@ -174,7 +174,7 @@ def prepare_input(self, input: "StandardInput") -> str:
         Returns:
             The prompt to send to the LLM.
         """
-        if self._pipeline.tokenizer.chat_template:  # type: ignore
+        if self._pipeline.tokenizer.chat_template is None:  # type: ignore
             return input[0]["content"]
 
         prompt: str = (

tests/unit/llms/huggingface/test_transformers.py

@@ -40,6 +40,21 @@ def test_model_name(self, transformers_llm: TransformersLLM) -> None:
             == "distilabel-internal-testing/tiny-random-mistral"
         )
 
+    def test_prepare_input(self, transformers_llm: TransformersLLM) -> None:
+        assert (
+            transformers_llm.prepare_input([{"role": "user", "content": "Hello"}])
+            == "<s> [INST] Hello [/INST]"
+        )
+
+    def test_prepare_input_no_chat_template(
+        self, transformers_llm: TransformersLLM
+    ) -> None:
+        transformers_llm._pipeline.tokenizer.chat_template = None
+        assert (
+            transformers_llm.prepare_input([{"role": "user", "content": "Hello"}])
+            == "Hello"
+        )
+
     def test_generate(self, transformers_llm: TransformersLLM) -> None:
         responses = transformers_llm.generate(
             inputs=[

tests/unit/steps/argilla/test_preference.py

@@ -83,13 +83,23 @@ def test_process(self, mock_dataset) -> None:
         )
         with patch.object(PreferenceToArgilla, "load"):
             step.load()
+
         step._instruction = "instruction"
         step._generations = "generations"
+        step._ratings = "ratings"
+        step._rationales = "rationales"
         step._dataset = mock_dataset  # type: ignore
 
         step._dataset.records.log = lambda x: x  # type: ignore
         assert list(
-            step.process([{"instruction": "test", "generations": ["test", "test"]}])
+            step.process(
+                [
+                    {
+                        "instruction": "test",
+                        "generations": ["test", "test"],
+                    }
+                ]
+            )
         ) == [[{"instruction": "test", "generations": ["test", "test"]}]]
         assert step._dataset.records  # type: ignore
 

tests/unit/steps/tasks/structured_outputs/test_outlines.py

@@ -101,6 +101,9 @@ class DummyUserTest(BaseModel):
 }
 
 
+@pytest.mark.skip(
+    reason="won't work until we update our code to work with `outlines>0.1.0`"
+)
 class TestOutlinesIntegration:
     @pytest.mark.parametrize(
         "format, schema, prompt",