Content Deep Dive

LLMON: An LLM-Native Markup Language, now in Granite 4.1 3B and 8B

Blog post from Hugging Face

Post Details
Company
Date Published
Author
Abraham Daniels, Basel Shbita, Chad DeLuca, Dan Gutfreund, Michael Hind, and Shub A
Word Count
2,022
Company Posts That Month
60
Language
-
Hacker News Points
-
Post removed?
No
Summary

IBM has introduced LLMON, or LLM Object Notation, an experimental XML-like markup language and two Granite 4.1 preview models designed to make context structure explicit for language models. LLMON separates instructions, data, requirements, and outputs into tagged, uniquely addressable blocks connected through execution directives, aiming to reduce confusion caused by flattened prompts and JSON’s parser-oriented syntax. IBM reports that the approach substantially improves resistance to static indirect prompt injection attacks by teaching models to treat unbound data and embedded instructions as nonauthoritative, while maintaining performance comparable to standard Granite 4.1 Instruct models on general benchmarks. The 3B and 8B preview models include chat templates that automatically render structured message components into LLMON, and accompanying tools can losslessly convert JSON to and from the format. LLMON is also extensible through custom tags and LoRA fine-tuning, demonstrated with ScienceWorld planning tasks and tool-calling workflows. IBM has released the models, code repository, conversion tools, notebooks, and an attention visualization demo for evaluation, while emphasizing that the technology remains an experimental preview rather than a production-ready system.

Trends Found in this Post
Trend Post Mentions Total Month Mentions Posts Companies MoM
LLM 10 3,355 742 191 -49%
AI Model Fine-tuning 2 613 130 56 -29%
Reinforcement learning 2 99 55 28 -39%
RAG 1 350 93 53 -60%
Use This Data

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.