---
license: other
language:
- en
base_model:
- unsloth/Qwen2.5-Coder-14B-Instruct-bnb-4bit
pipeline_tag: text-generation
tags:
- code
- reasoning
- local-ai
- qwen
- unsloth
- math
---

# 🧠 Medhavi-14B-Systems-Reasoning (GGUF)

**Medhavi-14B** is a high-density reasoning model fine-tuned for **low-level systems mastery, architectural analysis, and production-grade code generation.** Unlike standard chat models, Medhavi is trained to act as a Lead Systems Architect, performing an exhaustive "Deep Think" phase to analyze time/space complexity, edge cases, and low-level implementation details before outputting a finalized solution.

## ⚡ Inference Structure (Reasoning Chain)
The model follows a strict execution pipeline:
1. **System Prompt:** You are Medhavi, an expert reasoning and coding AI. For every instruction, you must perform a deep architectural and logical analysis inside <think> tags before providing the final output. Always end your response with [DONE].
2. **Context** 32K is supported but for best results use 20480

## 💾 Available Quantizations
* **`Q4_K_M` (Standard):** Ideal for local systems with 16GB RAM.
* **`Q6_K` (High Precision):** Recommended for complex coding tasks where 99% logic retention is required.

## 🚀 How to Run (LM Studio / Local)
This model requires a specific prompt format to trigger the reasoning circuits correctly.