Skip to content

Yoons-AI-LAB/Nexus-VMC-v1.4-RCL

Repository files navigation

Nexus-VMC-v1.4-RCL (1.5B)

Unified Strategic Intelligence for Medical & Enterprise Reasoning

Developed by Yoons-AI-LAB, Nexus-VMC-v1.4-RCL is a high-performance Small Language Model (SLM) engineered for the Antigravity Hub ecosystem. It delivers "Large Model" reasoning integrity and tool-calling precision in a compact 1.5B parameter architecture.


🚀 Vision

The goal of the Nexus evolution is to prove that high-integrity reasoning (Thinking-Chain) and complex functional calling can be internalized into ultra-lightweight models reachable on consumer hardware (8GB RAM).

📊 Performance Leaderboard (v1.4-RCL)

Rank Model Name Domain Integrity (Think/Tool) Speed (TPS)
🥇 1 Nexus-VMC-v1.4-RCL Unified (Med/Excel) 100% / 100% 130.3
🥈 2 hauhau-qwen9b General/Large 50% / 0% 37.2
🥉 3 phi3.5:latest General/SLM 0% / 0% 94.1

Benchmarked on 4070 Laptop GPU via Ollama API. TPS = Tokens Per Second.

✨ Key Features

  • VMC-RCL Protocol: Native support for reasoning-first execution using <vmc_think> blocks followed by [TOOL_CALL] JSON integration.
  • Medical Reasoning: Fine-tuned on specialized clinical scenarios and hospital resource management logic.
  • Excel Master: Accurate generation of complex spreadsheet logic, including SUMIF and data cleaning formulas.
  • Efficient Deployment: Optimized for Ollama/llama.cpp with a high-fidelity Q8_0 GGUF quantization.

📦 Getting Started

1. Download Model

The models are hosted on Hugging Face: 🔗 nebada1101/Nexus-VMC-v1.4-RCL

2. Run with Ollama

# 1. Download the .gguf from Hugging Face
# 2. Create the Modelfile provided in this repo
ollama create nexus-v1.4-rcl -f Modelfile
ollama run nexus-v1.4-rcl

🛠️ Repository Contents

This repository contains the core infrastructure used to develop and validate the model:

  • nexus_standardizer.py: Tool for normalizing HF weights for GGUF conversion.
  • weight_cleaner.py: Strips non-standard quantization artifacts from safetensors.
  • benchmark.py: The official Hub multi-domain benchmarking suite.
  • hf_uploader.py: automated deployment script for the Hugging Face Hub.

📄 License

Apache 2.0 - Developed by Yoons-AI-LAB.


© 2026 Yoons-AI-LAB. All Rights Reserved.

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages