File size: 2,472 Bytes
44388e3 0f3920c 44388e3 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 |
---
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:
```python
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. |