Flan T5 Base Samsum

🚀 flan-t5-base-samsum

This model is a fine - tuned version of google/flan-t5-base on the samsum dataset. It offers a powerful solution for text summarization, leveraging the pre - trained capabilities of the base model and adapting them to the specific characteristics of the samsum dataset.

🚀 Quick Start

How to use

from transformers import pipeline

pipe = pipeline("summarization", model="sharmax-vikas/flan-t5-base-samsum")

res = pipe('''dialogue: 
Margaret: Hi, in December I'd like to meet on 4th and 11th around 10:00 or 11:00. 
Evans: Hi, 4th - we can meet at 10:00.
Evans: And 11th - at 11:00. 
Margaret: Okey. And what about 18th?
Evans: I'm not sure about 18th. 
Evans: I might not be in town. 
Margaret: Okey, so we'll see. 
Evans: Yes. And I'll let you know next week. 
Margaret: If it's not 18th, maybe we could meet on 17th?
Evans: If I go away, I won't also be 17th.
Margaret: Okey, I get it. 
Evans: But we could meet 14th, if you like?
Margaret: Hm, I'm not sure whether I'm avaliable. 
Evans: So let's set these dates later, ok?
Margaret: Okey and we see each other 4th 10:00. 
Evans: Yes!''')

print(f"flan-t5-base summary:\n{res[0]['summary_text']}")


#output : flan-t5-base summary:
Margaret and Evans will meet on the 4th and 11th of December. They will meet at 10:00 on the 18th and at 11:00 on the 17th. If it's not 18th, they can meet on 17th or 14th.

✨ Features

This fine - tuned model achieves the following results on the evaluation set:

• Loss: 1.3736
• Rouge1: 47.355
• Rouge2: 23.7601
• Rougel: 39.8403
• Rougelsum: 43.4718
• Gen Len: 17.1575

🔧 Technical Details

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e - 05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon = 1e - 08
• lr_scheduler_type: linear
• num_epochs: 5

Training results

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.0
• Datasets 3.0.0
• Tokenizers 0.19.1

📄 License

This project is licensed under the Apache - 2.0 license.