feat: HuggingFaceLocalGenerator- allow passing generation_kwargs in run method (#6220)

* allow custom generation_kwargs in run

* reno

* make pylint ignore too-many-public-methods

haystack/preview/components/generators/hugging_face/hugging_face_local.py

@@ -206,14 +206,24 @@ class HuggingFaceLocalGenerator:
         )
 
     @component.output_types(replies=List[str])
-    def run(self, prompt: str):
+    def run(self, prompt: str, generation_kwargs: Optional[Dict[str, Any]] = None):
+        """
+        Run the text generation model on the given prompt.
+
+        :param prompt: A string representing the prompt.
+        :param generation_kwargs: Additional keyword arguments for text generation.
+        :return: A dictionary containing the generated replies.
+        """
         if self.pipeline is None:
             raise RuntimeError("The generation model has not been loaded. Please call warm_up() before running.")
 
         if not prompt:
             return {"replies": []}
 
-        output = self.pipeline(prompt, stopping_criteria=self.stopping_criteria_list, **self.generation_kwargs)
+        # merge generation kwargs from init method with those from run method
+        updated_generation_kwargs = {**self.generation_kwargs, **(generation_kwargs or {})}
+
+        output = self.pipeline(prompt, stopping_criteria=self.stopping_criteria_list, **updated_generation_kwargs)
         replies = [o["generated_text"] for o in output if "generated_text" in o]
 
         if self.stop_words:

releasenotes/notes/hflocalgenerator-generation-kwargs-in-run-2bde10d398a3712a.yaml

@@ -0,0 +1,5 @@
+---
+preview:
+  - |
+    Allow passing `generation_kwargs` in the `run` method of the `HuggingFaceLocalGenerator`.
+    This makes this common operation faster.

test/preview/components/generators/hugging_face/test_hugging_face_local_generator.py

@@ -1,3 +1,4 @@
+# pylint: disable=too-many-public-methods
 from unittest.mock import patch, Mock
 
 import pytest
@@ -242,6 +243,23 @@ class TestHuggingFaceLocalGenerator:
 
         assert results == {"replies": []}
 
+    @pytest.mark.unit
+    def test_run_with_generation_kwargs(self):
+        generator = HuggingFaceLocalGenerator(
+            model_name_or_path="google/flan-t5-base",
+            task="text2text-generation",
+            generation_kwargs={"max_new_tokens": 100},
+        )
+
+        # create the pipeline object (simulating the warm_up)
+        generator.pipeline = Mock(return_value=[{"generated_text": "Rome"}])
+
+        generator.run(prompt="irrelevant", generation_kwargs={"max_new_tokens": 200, "temperature": 0.5})
+
+        generator.pipeline.assert_called_once_with(
+            "irrelevant", max_new_tokens=200, temperature=0.5, stopping_criteria=None
+        )
+
     @pytest.mark.unit
     def test_run_fails_without_warm_up(self):
         generator = HuggingFaceLocalGenerator(