@@ -11,8 +11,8 @@ description: |
1111 - **Layout preservation** — tables, lists, and nested structures are converted to Markdown
1212 - **Multi-format support** — PDF, DOCX, PPTX, HTML, and images
1313 - **Better RAG quality** — structured chunks with heading metadata produce more relevant retrieval results
14- - **VLM-based processing ** — optionally route Vision Language Model inference through the stack's model-serving
15- infrastructure for richer document understanding (layout analysis, OCR via vision models)
14+ - **VLM configuration validation ** — VLM processing requires worker-side inference routing, which is not yet
15+ supported; configuring `vlm_model` currently produces a clear startup error
1616
1717 ## Usage
1818
@@ -33,23 +33,13 @@ description: |
3333 config: {}
3434 ```
3535
36- ### Enabling VLM Processing
36+ ### VLM Processing
3737
38- To enable VLM-based document processing, set `vlm_model` to a vision model registered with the
39- stack's inference API. The VLM pipeline routes inference through the stack's model-serving
40- infrastructure — no separate GPU resources are needed for document processing .
38+ VLM-based document processing is not currently available while this provider runs in worker
39+ mode because auto-routed inference cannot yet be reconstructed in the worker process. A
40+ configuration with `vlm_model` fails startup rather than silently running a different pipeline .
4141
42- ```yaml
43- file_processors:
44- - provider_id: docling
45- provider_type: inline::docling
46- config:
47- vlm_model: granite-docling-258M
48- vlm_preset: granite_docling
49- ```
50-
51- When `vlm_model` is not set or no inference provider is available, the processor gracefully
52- degrades to the standard non-VLM pipeline.
42+ Leave `vlm_model` unset to use the standard non-VLM pipeline.
5343
5444 ## Installation
5545
@@ -80,8 +70,8 @@ preserves semantic boundaries. It supports PDF, DOCX, PPTX, HTML, and images.
8070- ** Layout preservation** — tables, lists, and nested structures are converted to Markdown
8171- ** Multi-format support** — PDF, DOCX, PPTX, HTML, and images
8272- ** Better RAG quality** — structured chunks with heading metadata produce more relevant retrieval results
83- - ** VLM-based processing ** — optionally route Vision Language Model inference through the stack's model-serving
84- infrastructure for richer document understanding (layout analysis, OCR via vision models)
73+ - ** VLM configuration validation ** — VLM processing requires worker-side inference routing, which is not yet
74+ supported; configuring ` vlm_model ` currently produces a clear startup error
8575
8676## Usage
8777
@@ -102,23 +92,13 @@ file_processors:
10292 config : {}
10393` ` `
10494
105- ### Enabling VLM Processing
95+ ### VLM Processing
10696
107- To enable VLM-based document processing, set ` vlm_model` to a vision model registered with the
108- stack's inference API. The VLM pipeline routes inference through the stack's model-serving
109- infrastructure — no separate GPU resources are needed for document processing.
110-
111- ` ` ` yaml
112- file_processors:
113- - provider_id: docling
114- provider_type: inline::docling
115- config:
116- vlm_model: granite-docling-258M
117- vlm_preset: granite_docling
118- ` ` `
97+ VLM-based document processing is not currently available while this provider runs in worker
98+ mode because auto-routed inference cannot yet be reconstructed in the worker process. A
99+ configuration with ` vlm_model` fails startup rather than silently running a different pipeline.
119100
120- When `vlm_model` is not set or no inference provider is available, the processor gracefully
121- degrades to the standard non-VLM pipeline.
101+ Leave `vlm_model` unset to use the standard non-VLM pipeline.
122102
123103# # Installation
124104
0 commit comments