Merge branch 'develop' of https://github.com/argilla-io/distilabel in…

…to fix-exam-example

src/distilabel/models/llms/azure.py

@@ -122,7 +122,8 @@ def load(self) -> None:
         # This is a workaround to avoid the `OpenAILLM` calling the _prepare_structured_output
         # in the load method before we have the proper client.
         with patch(
-            "distilabel.models.openai.OpenAILLM._prepare_structured_output", lambda x: x
+            "distilabel.models.llms.openai.OpenAILLM._prepare_structured_output",
+            lambda x: x,
         ):
             super().load()
 

src/distilabel/models/llms/base.py

@@ -476,7 +476,7 @@ def _prepare_kwargs(
         Args:
             arguments: The arguments that would be passed to the LLM as **kwargs.
                 to update with the structured output configuration.
-            structured_outputs: The structured output configuration to update the arguments.
+            structured_output: The structured output configuration to update the arguments.
 
         Returns:
             kwargs updated with the special arguments used by `instructor`.

src/distilabel/models/llms/vllm.py

@@ -243,7 +243,10 @@ def _cleanup_vllm_model(self) -> None:
 
         destroy_model_parallel()
         destroy_distributed_environment()
-        del self._model.llm_engine.model_executor
+
+        # Don't delete model_executor if it does not exist, e.g. when VLLM_USE_V1 is set
+        if hasattr(self._model.llm_engine, "model_executor"):
+            del self._model.llm_engine.model_executor
         del self._model
         with contextlib.suppress(AssertionError):
             torch.distributed.destroy_process_group()

src/distilabel/pipeline/base.py

@@ -734,7 +734,7 @@ def _set_runtime_parameters(self, parameters: Dict[str, Dict[str, Any]]) -> None
 
         Args:
             parameters: A dictionary with the step name as the key and a dictionary with
-            the parameter name as the key and the parameter value as the value.
+                the parameter name as the key and the parameter value as the value.
         """
         step_names = set(self.dag.G)
         for step_name, step_parameters in parameters.items():

tests/integration/test_embedding_dedup.py

@@ -103,7 +103,7 @@ def test_embedding_deduplication() -> None:
 
         # NOTE: Guide to choose an index: https://github.com/facebookresearch/faiss/wiki/Guidelines-to-choose-an-index
         nn = FaissNearestNeighbour(
-            k=3,
+            k=5,
             metric_type=faiss.METRIC_INNER_PRODUCT,
             search_batch_size=50,
             # string_factory="IVF300_HNSW32,Flat",
@@ -122,8 +122,8 @@ def test_embedding_deduplication() -> None:
 
     ds = distiset["default"]["train"]
     ds_dedup = ds.filter(lambda x: x["keep_row_after_embedding_filtering"])
-    print(len(ds_dedup))
-    assert len(ds_dedup) == 71
+
+    assert len(ds_dedup) == 63
 
 
 if __name__ == "__main__":