---
license: mit
datasets:
- openai/gsm8k
base_model:
- Qwen/Qwen3-0.6B
tags:
- reasoning
---
# 🧠 Qwen-0.6B Reasoning – XformAI Fine-Tuned Model

**Model:** `XformAI-india/qwen-0.6b-reasoning`  
**Base Model:** [`Qwen/Qwen3-0.6B`](https://huggingface.co/Qwen/Qwen3-0.6B)  
**Architecture:** Transformer decoder (GPT-style)  
**Fine-Tuned By:** [XformAI](https://xformai.in)  
**Release Date:** May 2025  
**License:** MIT

---

## 🧠 What is it?

`qwen-0.6b-reasoning` is a **compact transformer model fine-tuned for reasoning, logic, and analytical thinking**.

Despite its size, it demonstrates strong performance across:
- 🧩 Riddles & Puzzles  
- 🧮 Math Word Problems  
- 🧠 Symbolic Reasoning  
- 💬 Chain-of-Thought Prompting  
- 🔍 Common Sense Logic  

> Fine-tuned on a curated instruction-style dataset focused on multi-step reasoning.

---

## 🚀 Why it Matters

- Performs like a **7B model** on reasoning benchmarks  
- **Lightweight (600M)** and can run on CPU or mobile edge devices  
- Excels in **step-by-step explanations** and **problem solving**

---

## 🧪 Fine-Tuning Overview
 ----------------------------------------------------------
| Category             | Detail                           |
|----------------------|----------------------------------|
| Base Model           | Qwen 0.6B                        |
| Target Objective     | Reasoning, logic, CoT            |
| Fine-Tuning Type     | Instruction                      |
| Optimizer            | AdamW (LoRA tuning)              |
| Precision            | bfloat16                         |
| Epochs               | 2                                |
| Max Tokens           | 2048                             |

---

## 🧩 Prompt Example

```python
from transformers import AutoTokenizer, AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained("XformAI-india/qwen-0.6b-reasoning")
tokenizer = AutoTokenizer.from_pretrained("XformAI-india/qwen-0.6b-reasoning")

prompt = "A farmer has 17 sheep. All but 9 run away. How many are left?"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))