lagna360/yo-router-270m

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.3BQuant:BF16Context Size:32kPublished:Aug 20, 2026License:gemmaArchitecture:Transformer Featherless Exclusive Cold

lagna360/yo-router-270m is a 270M-parameter tool router for macOS, fine-tuned from Google's FunctionGemma-270m-it. This model specializes in taking a single plain-English sentence and emitting a single function call, with its 10 macOS-specific tool declarations baked directly into the weights for a 20x reduction in prompt tokens. It achieves 95.2% tool accuracy and 82.3% argument accuracy, making it highly efficient for routing macOS system queries to predefined read-only tools.

Loading preview...

Overview

lagna360/yo-router-270m is a specialized 270M-parameter tool router designed for macOS. It is a LoRA fine-tune of google/functiongemma-270m-it, with its 10 macOS-specific tool declarations fused directly into the weights. This unique approach reduces prompt tokens from 655 to just 33, offering a significant efficiency gain.

Key Capabilities & Performance

This model excels at classifying plain-English sentences into one of 10 predefined read-only macOS system queries and extracting typed arguments. It achieves 95.2% tool accuracy and 82.3% argument accuracy on a hand-written test set of 249 cases. This represents a substantial improvement over the un-tuned functiongemma-270m-it (56.2% tool accuracy). The model is designed to be a single stage in a three-stage pipeline: utterance → ROUTER (this model) → EXECUTOR → RENDERER, ensuring structured and safe execution.

The 10 Baked-in Tools

The model's functionality is limited to these read-only macOS system queries:

  • disk_usage: Identifies folders consuming space.
  • storage_summary: Reports available storage.
  • largest_files: Lists biggest individual files.
  • find_files: Locates files by name.
  • top_processes: Shows processes by CPU/RAM usage.
  • network_listeners: Identifies what's using a port.
  • network_info: Provides IP, Wi-Fi SSID, gateway.
  • battery_status: Reports charge level and plug-in status.
  • system_info: Displays macOS version, host, user, uptime.
  • datetime: Returns current time and date.

Unique Differentiators

  • Baked-in Tool Declarations: Tools are part of the model weights, drastically reducing prompt size and improving inference speed.
  • High Accuracy for Narrow Domain: Achieves high accuracy for its specific task of macOS tool routing, outperforming its base model significantly.
  • Efficiency: Optimized for local execution on Apple Silicon, with a p50 latency of 0.238s on an M1 Pro.
  • Safety-Oriented Design: Intended for use in a pipeline where its output is treated as untrusted input, with execution handled by a separate, secure executor.

Limitations and Considerations

  • No General-Purpose Chat: Not suitable for general chat, reasoning, or text generation.
  • Immutable Tool Set: Adding, removing, or changing tools requires retraining.
  • macOS-Specific: Tools are designed for macOS platforms only.
  • Argument Accuracy: While high, 82.3% means arguments should be validated before execution.
  • Prompt-Shape Brittleness: Requires a specific prompt format without explicit tool declarations.
  • Licensing: This model is a derivative of Google's FunctionGemma and is subject to the Gemma Terms of Use, not an OSI-approved open-source license.

Finetunes
1 models