metadata

language:
  - en
tags:
  - qwen2
  - text-generation
  - summarization
  - key-points
  - blog-summarization
datasets:
  - ncls-p/blog-key-points
license: cc-by-4.0
base_model: Qwen/Qwen2.5-3B-Instruct

Qwen2.5-3B-blog-key-points

This model is fine-tuned from Qwen/Qwen2.5-3B-Instruct on the ncls-p/blog-key-points. It specializes in extracting key points from blog articles and web content, providing concise bullet-point summaries that capture the essential information.

Model Description

Qwen2.5-3B-blog-key-points is a 3B parameter model fine-tuned specifically for the task of extracting key points from articles. It can process a full article and generate a concise, bullet-point summary highlighting the most important information.

Model Details

Model Type: Qwen2.5 (3B parameters)
Base Model: Qwen/Qwen2.5-3B-Instruct
Training Dataset: ncls-p/blog-key-points
Language: English
License: CC-BY-4.0
Finetuning Approach: Instruction fine-tuning on article-summary pairs

Uses

Direct Use

This model is designed for extracting key points from articles. You can use it directly for:

Summarizing blog posts
Extracting important information from news articles
Creating bullet-point summaries of long-form content
Generating concise overviews of research papers

Example Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "ncls-p/Qwen2.5-3B-blog-key-points"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

article = """
[Your article text here]
"""

prompt = f"""
Extract the key points from the following article:

{article}
"""

inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=1024)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)

print(response)

Training

The model was fine-tuned on the blog-key-points dataset, which contains 200 article-summary pairs. Each pair consists of a full article and a bullet-point summary of key points extracted using AI.

Training Procedure

Fine-tuning Framework: Unsloth

Training Data Format:

{
  "instruction": "",
  "input": "Full article content",
  "output": "Here are the key points of the article:\n* Key point 1\n* Key point 2\n* Key point 3\n..."
}

Evaluation

The model was evaluated on its ability to extract relevant key points from articles not seen during training. Evaluation metrics focused on:

Relevance: How well the extracted points capture the main ideas of the article
Conciseness: The ability to summarize information in a clear, bullet-point format
Completeness: Whether all important information is captured in the summary

Limitations and Biases

The model may inherit biases present in the training data, including potential biases in the source articles or in the key point extraction process.
Performance may vary depending on the length, complexity, and domain of the input article.
The model is primarily trained on English-language content and may not perform well on content in other languages.
As with any summarization model, there is a risk of omitting important information or misrepresenting the original content.

How to Cite

If you use this model in your research, please cite:

@misc{qwen25-3b-blog-key-points,
  author = {ncls-p},
  title = {Qwen2.5-3B-blog-key-points},
  year = {2024},
  publisher = {Hugging Face},
  journal = {Hugging Face model repository},
  howpublished = {\url{https://huggingface.co/ncls-p/Qwen2.5-3B-blog-key-points}},
}

Dataset Creation

The dataset used to train this model was created using the llm-to-blog-key-points-dataset, a CLI tool that extracts key points from web articles using AI and adds them to a dataset in a structured format.