bluey-poor-flant5

Instruction Fine-tuned FLAN-T5 Large for Text Transformation with implicit Grammar and Spelling Correction, dubbed Bleuy-poor

I am GPU poor, this instruction fine tuned model is only fine tuned for 3 epochs, for $10 dollars two A100 GPUs 80GB rented for 9 hours.

This model is a instruction fine-tuned version of FLAN-T5 Large for text transformation tasks on https://huggingface.co/datasets/sugiv/synthetic-text-transformation-dataset/viewer/default/train

Usage

# Load model and tokenizer
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
from peft import PeftModel, PeftConfig

# Load the PEFT configuration
peft_model_id = "sugiv/bluey-poor-flant5"
peft_config = PeftConfig.from_pretrained(peft_model_id)

# Load the base model
base_model = AutoModelForSeq2SeqLM.from_pretrained(peft_config.base_model_name_or_path)

# Load the PEFT model
model = PeftModel.from_pretrained(base_model, peft_model_id)

# Load the tokenizer
tokenizer = AutoTokenizer.from_pretrained(peft_config.base_model_name_or_path)

# Set the model to evaluation mode
model.eval()

def generate_transform_prompt(input_text, filter_combination):
    return f'''You are an advanced text transformation AI. Your task is to {filter_combination['Task']} the given input text according to the specified parameters. {filter_combination['Task'].capitalize()}ing means expressing the same meaning using different words, while maintaining the original intent. Always correct spelling and grammatical errors implicitly.

User: Transform the following text based on these parameters:
Task: {filter_combination['Task']}
Tone: {filter_combination['Tone']}
Target Audience: {filter_combination['Target Audience']}
Complexity: {filter_combination['Complexity']}
Purpose: {filter_combination['Purpose']}
Style: {filter_combination['Style']}
Verbosity: {filter_combination['Verbosity']}

Input Text: {input_text}

Instructions:
1. {filter_combination['Task']} the text according to the specified parameters.
2. Maintain the original meaning, context, jargon, and entities.
3. Adjust the language complexity and verbosity as specified.
4. Optimize the text for the target audience and purpose.
5. Ensure the output is coherent and flows naturally.
6. Implicitly correct any spelling or grammatical errors.

Transformed text:'''

# Example usage
input_text = "The quick brown fox jumps over the lazy dog."
filter_combination = {
    "Task": "Rephrase",
    "Tone": "Professional",
    "Target Audience": "Business executives",
    "Complexity": "Advanced",
    "Purpose": "Inform",
    "Style": "Analytical",
    "Verbosity": "Concise"
}

prompt = generate_transform_prompt(input_text, filter_combination)
inputs = tokenizer(prompt, return_tensors="pt", max_length=512, truncation=True)
outputs = model.generate(**inputs, max_length=150, num_return_sequences=1)
transformed_text = tokenizer.decode(outputs, skip_special_tokens=True)
print(transformed_text)