feat: workers without shared models and convert params (#304)

Signed-off-by: Michele Dolfi <dol@zurich.ibm.com>

Author: dolfim-ibm

docling_serve/orchestrator_factory.py

@@ -19,6 +19,7 @@ def get_async_orchestrator() -> BaseOrchestrator:
 
         local_config = LocalOrchestratorConfig(
             num_workers=docling_serve_settings.eng_loc_num_workers,
+            shared_models=docling_serve_settings.eng_loc_share_models,
         )
 
         cm_config = DoclingConverterManagerConfig(

docling_serve/settings.py

@@ -63,6 +63,7 @@ class DoclingServeSettings(BaseSettings):
     eng_kind: AsyncEngine = AsyncEngine.LOCAL
     # Local engine
     eng_loc_num_workers: int = 2
+    eng_loc_share_models: bool = False
     # KFP engine
     eng_kfp_endpoint: Optional[AnyUrl] = None
     eng_kfp_token: Optional[str] = None

docs/configuration.md

@@ -66,6 +66,7 @@ The following table describes the options to configure the Docling Serve local e
 | ENV | Default | Description |
 |-----|---------|-------------|
 | `DOCLING_SERVE_ENG_LOC_NUM_WORKERS` | 2 | Number of workers/threads processing the incoming tasks. |
+| `DOCLING_SERVE_ENG_LOC_SHARE_MODELS` | False | If true, each process will share the same models among all thread workers. Otherwise, one instance of the models is allocated for each worker thread. |
 
 #### KFP engine
 

pyproject.toml

@@ -35,7 +35,7 @@ requires-python = ">=3.10"
 dependencies = [
     "docling~=2.38",
     "docling-core>=2.44.1",
-    "docling-jobkit[kfp,vlm]~=1.2",
+    "docling-jobkit[kfp,vlm]~=1.3",
     "fastapi[standard]~=0.115",
     "httpx~=0.28",
     "pydantic~=2.10",