AI BriefingPhocinae-Largha-150M-v1: 144M Parameter Local Decision Engine Reduces LLM Calls by 55%
Phocinae/Phocinae-Largha-150M-v1
About the project
Handles structured decisions such as approvals, tool selection, and step validation that recur in agent sessions. Given a state and a list of questions as input, it instantly returns answers with confidence scores in a single forward pass.
A lightweight model with 144.3M parameters, it achieves low latency of 21ms on an RTX 5090. By routing only low-confidence decisions to a higher-tier LLM, it reduces API call volume by 55% while maintaining accuracy.
Unlike open-source LLMs, it is not suitable for chat or long-document reasoning. It is useful for security-sensitive environments requiring a local decision gateway without data leakage, or for agent workflows needing cost optimization.
Phocinae/Phocinae-Largha-150M-v1
The original page has no description.
text-classification
This introduction was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report errors, attribution issues, or removal requests via Contact.