flan-t5-finetuned1

Model Card: FLAN-T5 Fine-Tuned on CNN/DailyMail 3.0

Model Details Model Name: FLAN-T5 Fine-Tuned on CNN/DailyMail 3.0 Model Type: Sequence-to-Sequence Transformer (T5) Base Model: FLAN-T5 (google/flan-t5-base) Fine-Tuned On: CNN/DailyMail 3.0 dataset Architecture: T5 (Text-to-Text Transfer Transformer) Model Size: Base version (Google FLAN-T5) Training Framework: Transformers and Datasets from Hugging Face Training Data Dataset Used: CNN/DailyMail 3.0, which contains news articles and human-written summaries. Dataset Size: The CNN/DailyMail 3.0 dataset includes thousands of news articles, with highlights used as summaries. Training Details Hardware: Model trained on GPU (CUDA). Batch Size: 4 for both training and evaluation. Learning Rate: 2e-5. Epochs: 3. Evaluation Strategy: Evaluated after each epoch. Preprocessing: Input articles were truncated and padded to a max length of 512 tokens. Summaries (highlights) were truncated and padded to a max length of 150 tokens.

Usage How to Use To use this model for summarization:

python

from transformers import T5Tokenizer, T5ForConditionalGeneration

Load the fine-tuned model and tokenizer

tokenizer = T5Tokenizer.from_pretrained("anaam819/flan-t5-finetuned1") model = T5ForConditionalGeneration.from_pretrained("anaam819/flan-t5-finetuned1")

Example input text

text = "Input news article text here."

Preprocess the input and generate summary

inputs = tokenizer.encode(text, return_tensors="pt", max_length=512, truncation=True) summary_ids = model.generate(inputs, max_length=150, num_beams=4, early_stopping=True) summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)

print(summary)

Limitations The model may sometimes generate incomplete or irrelevant summaries if the input text is very lengthy or contains ambiguous information. Requires a GPU for efficient inference, especially on longer sequences. Ethical Considerations Summarization models can potentially generate biased or misleading summaries, depending on the input text. Users should verify the accuracy of the generated summaries before using them in critical applications.

Acknowledgements This model was trained using the FLAN-T5 architecture provided by Google, and fine-tuned using the CNN/DailyMail 3.0 dataset.