Benchmark Chunk Viewer

Compare stored chunks with the selected PDF processing pipeline and its page, table, chart, and vision settings.

Idle.

Processing options Docling

Docling timeout (seconds)

Docling threads

Docling max concurrency

Docling max assets

Docling vision concurrency

Force OCR empty-page ratio

Force OCR median chars

Quick vision default concurrency

Quick region vision concurrency

Quick page vision concurrency

Quick native layout backend

Quick native text threshold

Quick max table/chart regions

Disable page reconstruction

Save intermediates

Disable all quick vision

Route weak pages to full-page vision

Send tables/charts to vision

Allow OCR in PyMuPDF layout stage

Route image pages to full-page vision

Route visual candidates to full-page vision

Qwen context cap (tokens)

Qwen image/input reservation

Qwen maximum Markdown tokens

Qwen render DPI

Qwen max image pixels

OOM crop height (inches)

Enable Qwen thinking (slower)

Changes apply on the next load or processing action. Qwen is benchmark-only, uses one GPU request at a time, and retries memory/context failures as overlapping vertical crops.

Qwen model status unknown

Pages0

Stored chunks0

Parse ms0

Pages / sec0

PDF Processing Cost

Process all pages to collect per-page token usage and estimate total processing cost from a custom per-token price.

No summary loaded.

Cheap model input cost / token (USD)

Cheap model output cost / token (USD)

Chat model input cost / token (USD)

Chat model output cost / token (USD)

Total tokens0

Total cost$0.00

Pages processed0

Known pages0

Page	Model tier	Model	Prompt tokens	Completion tokens	Total tokens	Cost	Mode
Process the PDF to see page-level token usage.

PDF page

Select a document and page.

Reconstructed page markdown

Process one page at a time so page reconstruction failures stay local to the selected page.

unknown route unknown 0 warnings 0 tokens $0.00

Secondary inspector

Stored ES chunks

0 chunks

Pipeline-generated structured chunks

0 chunks