---
base_model: Qwen/Qwen3-0.6B
language:
- en
library_name: transformers
pipeline_tag: text-generation
tags:
- qwen
- qwen3
- gguf
- title-generation
- summarization
- finetune
license: apache-2.0
---

# Titlewen-GGUF

**Titlewen** is a small specialized title-generation model based on **Qwen3-0.6B**, fine-tuned to turn user messages into short, descriptive conversation titles.

This repository contains the **GGUF** version intended for lightweight local inference with runtimes such as `llama.cpp`.

## What it does

Titlewen takes an arbitrary user message and produces a concise title describing its main topic.

Examples:

| Input | Output |
| --- | --- |
| `Hey we are at the coffee shop, do you want anything?` | `Coffee Shop Chat` |
| `CUDA out of memory. Tried to allocate 20.00 MiB...` | `CUDA Memory Allocation Error` |
| `how is this possible? rtx 3060...` | `RTX 3060 Performance Analysis` |
| `Help me make the README a bit better` | `Improving README Clarity` |

The model is designed for applications such as:

- automatic chat/conversation naming
- message and thread titles
- support-ticket titles
- compact topic extraction
- short UI labels derived from user input

## Model details

- **Base model:** `Qwen/Qwen3-0.6B`
- **Task:** Short title generation
- **Architecture:** Decoder-only Transformer
- **Training method:** LoRA fine-tuning
- **Training dataset:** approximately 20,000 title-generation examples
- **Output style:** typically 2–6 words
- **Format:** GGUF
- **Primary languages:** English