app.py

model_name = "BidhanAcharya/fine-tuned-sentiment-analyzer"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name).to("cuda")

def prepare_inference_input(review, instruction="You are good at reviewing positive, negative sentiment.\n\n"):
    # Combine the instruction and input text into one string
    input_text = f"{instruction}### Input:\n{review}\n### Response:"
    return input_text


def analyze_sentiment(review):
    # Prepare the input for inference
    inference_input = prepare_inference_input(review)

    # Tokenize the input
    input_tensor = tokenizer([inference_input], return_tensors="pt", padding=True).to("cuda")

    # Generate the output
    output = model.generate(
        **input_tensor,
        max_new_tokens=128,
        use_cache=True,
        temperature=0.7,
        top_p=0.9
    )

    # Decode the output , the output is in the form of list 
    decoded_output = tokenizer.batch_decode(output, skip_special_tokens=True)[0]

    # Regular expressions to extract the first Input and Response sections
    input_pattern = r'### Input:\n(.*?)\n###'
    response_pattern = r'### Response:\n(.*?)\n###'

    # Extracting the  Input section
    input_match = re.search(input_pattern, decoded_output, re.DOTALL)
    # Extracting the  Response section
    response_match = re.search(response_pattern, decoded_output, re.DOTALL)

    # Combining the extracted input and response into a dictionary, Extract the group(1) only : because of token size the model may generate the same output multiple times
    extracted_data = {
        'Input': input_match.group(1).strip() if input_match else None,
        'Response': response_match.group(1).strip() if response_match else None
    }

    return extracted_data['Response']

# Create the Gradio interface
interface = gr.Interface(
    fn=analyze_sentiment,
    inputs=gr.Textbox(lines=2, placeholder="Enter your review/sentiment here"),
    outputs=gr.Textbox(label="Sentiment Analysis Result"),
    title="Sentiment Analysis",
    description="Enter a movie review to analyze its sentiment."
)

# Launch the interface
interface.launch()