DEDUPED Hot AI repos. Real use cases. What people build. Get the Monday 5

How to raise accuracy with hybrid LLM mode with Marker

Add the use_llm flag so an LLM merges tables across pages and fixes inline math and forms.

Short answer: Install with pip install marker-pdf; then configure your LLM backend as described in the README LLM services section; run marker_single /path/to/file.pdf on a hard document. All steps.

  • Data and analytics
  • Intermediate
  • 30 minutes
  • From the README
  • Claude Code
  • Codex
  • Cursor
  • Gemini CLI

cost. Each LLM call is billed to your provider account.

data. Page content is sent to the LLM service you configure; do not use it on confidential documents without checking.

Who it is forA developer feeding hard PDFs, with math or forms, into a knowledge base.

The problemDefault conversion can mangle multi-page tables and inline equations.

The outcomeA higher quality markdown file with fewer broken tables and formulas.

What you need first

  • Python 3.10+ and PyTorch
  • pip
  • A GPU is best, CPU and Apple Silicon also work in fast mode

Steps

  1. Install with pip install marker-pdf
  2. Configure your LLM backend as described in the README LLM services section
  3. Run marker_single /path/to/file.pdf on a hard document
  4. Add --use_llm to enable hybrid mode
  5. Add --redo_inline_math with it for the highest quality math conversion

Prompt for your agent

Using datalab-to/marker, configure an LLM service from its README, then run `marker_single /path/to/file.pdf --use_llm --redo_inline_math` on <my PDF> and compare the result with a run without the flag.

Caveat. The default LLM service is Gemini and needs its own API key.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for datalab-to/marker and "Raise accuracy with hybrid LLM mode"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer