README.md

---
title: Stack X Ultimate
emoji: 🔥
colorFrom: purple
colorTo: blue
sdk: gradio
sdk_version: 6.13.0
app_file: app.py
pinned: false
description: >-
  Open-source agentic model with tool calling. Deploy on your own GPU — no API
  costs.
tags:
- agentic
- tool-calling
- llm
- qwen
- open-source
- local-ai
license: apache-2.0
---
# Stack X Ultimate — Agentic Tool-Calling Model

<div align="center">

**Open-source agentic model that calls real tools. Deploy on your GPU — no API key required.**

[![Model](https://img.shields.io/badge/Model-Qwen2.5--Coder--3B--Instruct-blue)](https://huggingface.co/Qwen/Qwen2.5-Coder-3B-Instruct)
[![Quantization](https://img.shields.io/badge/Quantization-QLoRA%204--bit-purple)](https://github.com/QwenLM/Qwen2.5-Coder)
[![License](https://img.shields.io/badge/License-Apache%202.0-green)](LICENSE)

</div>

---

## What It Does

Stack X Ultimate is a fine-tuned Qwen2.5-Coder-3B-Instruct optimized for **agentic tool-calling workflows**:

- 🔢 **Calculator** — evaluates mathematical expressions
- 🕐 **Current Time** — returns live UTC timestamp
- 📁 **File Search** — glob-based file discovery in directory trees
- ⚡ **Command Execution** — runs shell commands and returns structured output

The model decides *when* to call tools and *how* to interpret results — mirroring how GPT-4 and Claude handle function calling, but running entirely on your infrastructure.

---

## Quick Start

### Try in Browser
Use the chat interface above — try the pre-loaded examples or type your own query.

### Run Locally

```bash
# Clone the model
git lfs install
git clone https://huggingface.co/my-ai-stack/Stack-X-Ultimate

# Run with llama.cpp
./llama.cpp -m ./Stack-X-Ultimate/unsloth.Q4_K_M.gguf \
  -n 8192 \
  --ctx-size 8192 \
  -p "You are a helpful AI assistant with tool calling."

# Or with transformers
python3 -c "
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained('my-ai-stack/Stack-X-Ultimate')
tok = AutoTokenizer.from_pretrained('my-ai-stack/Stack-X-Ultimate')
print('Model loaded!')
"
```

---

## Architecture

| Component | Detail |
|---|---|
| Base Model | Qwen/Qwen2.5-Coder-3B-Instruct |
| Fine-tune Method | QLoRA 4-bit (r=32, all 7 target modules) |
| Training Data | 27,000+ agentic tool-call examples |
| Sequence Length | 8,192 tokens |
| VRAM Required | ~6 GB (Q4_K_M GGUF) |
| Min. Hardware | Single V100 16GB or equivalent |

---

## Available Tools

| Tool | Description | Example |
|---|---|---|
| `calculator` | Evaluate math expressions | `1500 * 0.07 * 30` |
| `get_current_time` | Return current UTC time | — |
| `search_files` | Glob-based file search | `*.py` in `./src` |
| `run_command` | Execute shell commands | `git status` |

---

## Enterprise Deployment

Need tool integrations specific to your stack? Want to deploy inside your VPC?

📬 **[Contact Stack AI →](https://www.stack-ai.me/contact)**

- Custom tool integrations (APIs, databases, internal systems)
- VPC-isolated deployment (AWS / GCP / Azure)
- Air-gapped / on-prem installation
- Custom LoRA fine-tuning on your data

---

## License

Apache 2.0 — free for commercial and personal use.

---

*Stack AI — Sovereign AI Infrastructure. [huggingface.co/my-ai-stack](https://huggingface.co/my-ai-stack) · [stack-ai.me](https://www.stack-ai.me)*