Fix agentics components: add areduce_batch_size and atype parameters

- Add areduce_batch_size=100 to SemanticAggregator to prevent overwhelming LLM
- Add atype parameter to AG initialization in SyntheticDataGenerator
- Add null check for output_states in SyntheticDataGenerator
- Return empty DataFrame if no output_states generated

Co-authored-by: Alfio Gliozzo <gliozzo@us.ibm.com>

Author: Alfio Gliozzo

src/lfx/src/lfx/components/agentics/agenerate_component.py

@@ -18,7 +18,7 @@
 from lfx.schema.dataframe import DataFrame
 
 
-class AgenerateComponent(BaseAgenticComponent):
+class SyntheticDataGenerator(BaseAgenticComponent):
     """Generate synthetic data using either example data or a defined schema.
 
     This component creates realistic synthetic data by either:
@@ -69,10 +69,6 @@ class AgenerateComponent(BaseAgenticComponent):
         IntInput(
             name="batch_size",
             display_name="Number of Rows to Generate",
-            info=(
-                "Number of new synthetic rows to generate. When an Input Table is provided, "
-                "the generated rows are appended to the original data."
-            ),
             value=10,
             advanced=False,
         ),
@@ -106,12 +102,12 @@ async def aGenerate(self) -> DataFrame:  # noqa: N802
         if self.source:
             source = AG.from_dataframe(DataFrame(self.source))
             atype = source.atype
-            instructions = str(self.instructions) if self.instructions else "Generate similar data based on the examples provided."
-            instructions += "\nHere are examples to take inspiration from:\n" + str(source.states[:50])
+            instructions = str(self.instructions)
+            instructions += "\nHere are examples to take inspiration from" + str(source.states[:50])
         elif self.schema != []:
             schema_fields = build_schema_fields(self.schema)
             atype = create_pydantic_model(schema_fields, name="GeneratedData")
-            instructions = str(self.instructions) if self.instructions else "Generate realistic synthetic data following the provided schema."
+            instructions = str(self.instructions)
         else:
             msg = "Synthetic data generation requires either a sample DataFrame or schema definition (but not both)."
             raise ValueError(msg)
@@ -122,14 +118,9 @@ async def aGenerate(self) -> DataFrame:  # noqa: N802
             llm=llm,
             instructions=instructions,
         )
-        # Ensure output_states is a list, not None
-        if output_states is None:
-            output_states = []
-        
-        if self.source:
-            output_states = source.states + output_states
-        
-        output = AG(atype=atype, states=output_states)
-
-        return DataFrame(output.to_dataframe().to_dict(orient="records"))
-
+        if output_states:
+            if self.source:
+                output_states = source.states + output_states
+            output = AG(atype=atype, states=output_states)
+            return DataFrame(output.to_dataframe().to_dict(orient="records"))
+        return DataFrame([])

src/lfx/src/lfx/components/agentics/areduce_component.py

@@ -113,6 +113,7 @@ async def aReduce(self) -> DataFrame:  # noqa: N802
                 else "\nGenerate a list of instances of the target type following those instructions : ."
                 + self.instructions,
                 llm=llm,
+                areduce_batch_size=100,
             )
 
             output = await (target << source)