Added input/output type declaration.

llama_stack/apis/preprocessing/preprocessing.py

@@ -14,15 +14,19 @@ from llama_stack.apis.vector_io import Chunk
 from llama_stack.schema_utils import json_schema_type, webmethod
 
 
-class PreprocessingInputType(Enum):
-    document_content = "document_content"
-    document_path = "document_path"
+class PreprocessingDataType(Enum):
+    document_uri = "document_uri"
+    document_directory_uri = "document_directory_uri"
+
+    binary_document = "binary_document"
+    raw_text_document = "raw_text_document"
+    chunks = "chunks"
 
 
 @json_schema_type
 class PreprocessingInput(BaseModel):
     preprocessor_input_id: str
-    preprocessor_input_type: Optional[PreprocessingInputType]
+    preprocessor_input_type: Optional[PreprocessingDataType]
     path_or_content: str | URL
 
 

llama_stack/providers/inline/preprocessing/docling/docling.py

@@ -11,6 +11,7 @@ from docling_core.transforms.chunker.hybrid_chunker import HybridChunker
 from llama_stack.apis.common.content_types import URL
 from llama_stack.apis.preprocessing import (
     Preprocessing,
+    PreprocessingDataType,
     PreprocessingInput,
     PreprocessingResponse,
     Preprocessor,
@@ -22,6 +23,12 @@ from llama_stack.providers.inline.preprocessing.docling import InlineDoclingConf
 
 
 class InclineDoclingPreprocessorImpl(Preprocessing, PreprocessorsProtocolPrivate):
+    # this preprocessor receives URLs / paths to documents as input
+    INPUT_TYPES = [PreprocessingDataType.document_uri]
+
+    # this preprocessor either only converts the documents into a text format, or also chunks them
+    OUTPUT_TYPES = [PreprocessingDataType.raw_text_document, PreprocessingDataType.chunks]
+
     def __init__(self, config: InlineDoclingConfig) -> None:
         self.config = config
         self.converter = DocumentConverter()