Open
Description
Hi 👋
I'm currently using Marker with use_llm=True and a local Ollama setup. I'm experimenting with various models like deepseek-r1:14b, llama3, etc. for high-quality PDF conversion (scientific documents, multi-column, formulas, tables).
I wanted to ask:
- From your internal benchmarks, what LLMs perform best in hybrid mode (Marker + LLM)?
- Are there specific models (e.g. Gemini, Claude, Deepseek, OpenAI, etc.) that you've found significantly outperform others in table handling, math rendering, or section parsing?
- Any examples of LLM-judge or heuristic scores for
use_llm=Trueconfigurations?
Thanks a lot — I'm using Marker in a research context and want the most accurate setup, even if it's slow 🚀
Metadata
Metadata
Assignees
Labels
No labels