phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-08-07 02:58:21 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Added lazy initialization to the docling provider.

Browse source
This commit is contained in:
ilya-kolchinsky 2025-03-07 15:38:34 +01:00
parent 275fdbc23f
commit e895bb111c
1 changed files with 9 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

14
llama_stack/providers/inline/preprocessing/docling/docling.py
View file

@ -36,13 +36,10 @@ class InclineDoclingPreprocessorImpl(Preprocessing, PreprocessorsProtocolPrivate
def __init__(self, config: InlineDoclingConfig) -> None:
self.config = config
self.converter = DocumentConverter()
self.converter = None
self.chunker = None
async def initialize(self) -> None:
if self.config.chunk:
# TODO: docling should use Llama Stack's inference API instead of handling tokenization by itself
self.chunker = HybridChunker()
async def initialize(self) -> None: ...
async def shutdown(self) -> None: ...
@ -56,6 +53,13 @@ class InclineDoclingPreprocessorImpl(Preprocessing, PreprocessorsProtocolPrivate
preprocessor_inputs: List[PreprocessingDataElement],
options: Optional[PreprocessorOptions] = None,
) -> PreprocessorResponse:
if self.converter is None:
# this is the first time this method is called
self.converter = DocumentConverter()
if self.config.chunk and self.chunker is None:
# TODO: docling should use Llama Stack's inference API instead of handling tokenization by itself
self.chunker = HybridChunker()
results = []
for inp in preprocessor_inputs: