Zero-Shot Prompt Routing with Liquid AI's LFM 2.5-Encoder-350M

Melvin Vivas · X video post · 2026-08-27 · 2:16 · 839 views · Open on X

Topics: LLMOps, Deployment & Monitoring, AI System Design & Architecture, AI Agents, Tool Use & MCP · Level: intermediate

Summary

This demo shows how a small encoder model, Liquid AI's LFM 2.5-Encoder-350M, can route prompts zero-shot. It sends simple requests to cheap models and complex ones to expensive models. You set the categories at runtime, so you can add, remove or rewrite routes without retraining a classifier. The demo adds a "soccer agent" category on the fly, and the router sends soccer questions to it right away.

Key points

• Not every request needs the largest, most expensive model. Send complex tasks to frontier models and simple ones to smaller, cheaper (or free) models.
• Example routes for a device assistant: weather queries or setting a timer are simple function calls, while planning and booking a trip is a complex multi-step agentic task.
• Routing can also be done by domain, for example sending code prompts to the right programming-language handler, handling math separately, or classifying by topic.
• LFM 2.5-Encoder reads the full prompt and scores it against every category in a single forward pass. It runs locally and responds almost instantly.
• Categories are given at runtime, so you can add, remove or rewrite them without training a new classifier.
• Demo: 'Who won the 2026 FIFA World Cup?' first went to the complex agentic route. After a 'soccer agent' category was added at runtime, soccer questions went there, and 'set a timer' still routed correctly.
• The same mechanism can catch off-topic or low-value requests and send them to a cheaper model or a safe default, or decline them, so routing policy fits your use case.
• The creator argues that changing categories at runtime is the key feature for self-optimizing systems.

Resources mentioned

• LFM2.5-Encoder-350M · tool · huggingface.co · free
  A 350M-parameter encoder model from Liquid AI that does zero-shot prompt classification and routing against categories you supply at runtime.
  Also in: Zero-Shot Prompt Routing with Liquid AI's LFM 2.5-Encoder-350M (Melvin Vivas on X · notes), AIBackends v0.7.0: Prompt Routing with Liquid AI's LFM2.5 Encoder (Melvin Vivas on X · notes), Model Routing Fine-Tuned on Your Agent Harness Traces (Melvin Vivas on X · notes), Zero-Shot Prompt Routing by Task Complexity with LFM2.5-Encoder (Melvin Vivas on X · notes)
• Liquid AI (@liquidai) on X · website · x.com · free
  An AI company that builds efficient foundation models. The quoted post shows its PII handling working on Japanese text.
  Also in: Zero-Shot Prompt Routing with Liquid AI's LFM 2.5-Encoder-350M (Melvin Vivas on X · notes), Liquid AI LFM 2.5 Encoder: a CPU-friendly encoder model (Melvin Vivas on X · notes), Fine-tuning Liquid AI LFM2/LFM2.5 MoE models with the new Halo framework (Melvin Vivas on X · notes), Liquid AI's LFM2-Longevity models for aging-data analysis (Melvin Vivas on X · notes) and 19 more
• Original LFM 2.5-Encoder prompt routing demo video (YouTube) · video · youtube.com · free
  The full original YouTube demo of zero-shot prompt routing with LFM 2.5-Encoder. The caption link appears truncated and looked broken when checked.

Try this

• Route requests by complexity: send simple ones to small or cheap models and complex agentic tasks to the expensive models.
• Define routing categories at runtime and update them as you add new agents, with no retraining needed.
• Add routes that catch off-topic or low-value requests and send them to a cheaper model or a safe default, or decline them.
• Build a device-assistant router that sends simple function calls (weather, timers) to a small model and multi-step agentic tasks (trip planning and booking) to a larger model.
• Build a code router that sends each prompt to a handler or model for the right programming language.
• Add a new domain agent (for example a soccer agent) by adding its category to the router at runtime.