language: en
thumbnail: https://example.com/thumbnail.png
tags:
- paraphrasing
- T5
- text generation
- NLP
- transformers
license: mit
datasets:
- mteb/quora
metrics:
- accuracy
base_model:
- humarin/chatgpt_paraphraser_on_T5_base
library_name: transformers
ChatGPT and T5 Base Paraphraser
This model is a fine-tuned version of the T5 transformer model designed for paraphrasing questions using the ChatGPT architecture.
Model Description
The chat_gpt_and_t5_base_paraphraser model is trained to generate paraphrased versions of input questions by utilizing a sequence-to-sequence approach. The model leverages the T5 architecture and has been fine-tuned on the Quora Question-Answer dataset to improve its ability to create diverse and meaningful paraphrases.
Intended Use
This model is intended for applications where paraphrasing of text is required, such as:
- Chatbots
- Question-answering systems
- Content generation
- Educational tools
How to Use
To use the model, install the Hugging Face transformers library and follow these steps:
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
# Load the model and tokenizer
model_name = "jaesani/chat_gpt_and_t5_base_paraphraser"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
def paraphrase(question, max_length=128):
input_ids = tokenizer(f'paraphrase: {question}', return_tensors="pt", padding="longest", max_length=max_length, truncation=True).input_ids
outputs = model.generate(input_ids, max_length=max_length)
return tokenizer.decode(outputs[0], skip_special_tokens=True)
# Example usage
paraphrased_text = paraphrase("What are the best places to see in New York?")
print(paraphrased_text)
Training Data
The model was fine-tuned using the Quora Question-Answer Dataset, which consists of pairs of questions that may or may not be paraphrases of each other.
Evaluation
The model's performance can be evaluated based on the diversity and coherence of the paraphrases it generates. Specific metrics can include BLEU scores and human evaluations for semantic similarity.
Limitations
The model may produce paraphrases that are not contextually relevant. It may struggle with highly technical or domain-specific language. Generated paraphrases might be similar for closely related input questions.
License
This model is licensed under MIT License.