app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

# Load your fine-tuned model and tokenizer
model_name = "legacy107/flan-t5-large-bottleneck-adapter-cpgQA-unique"
tokenizer = AutoTokenizer.from_pretrained(model_name, device_map="auto")
model = AutoModelForSeq2SeqLM.from_pretrained(
    model_checkpoint, device_map="auto"
)
model.set_active_adapters("question_answering")

max_length = 512
max_target_length = 128

# Define your function to generate answers
def generate_answer(question, context):
    # Combine question and context
    input_text = f"question: {question} context: {context}"

    # Tokenize the input text
    input_ids = tokenizer(
        input_text,
        return_tensors="pt",
        padding="max_length",
        truncation=True,
        max_length=512,
    ).input_ids

    # Generate the answer
    with torch.no_grad():
        generated_ids = model.generate(input_ids, max_new_tokens=max_target_length)

    # Decode and return the generated answer
    generated_answer = tokenizer.decode(generated_ids[0], skip_special_tokens=True)

    return generated_answer

# Create a Gradio interface
iface = gr.Interface(
    fn=generate_answer,
    inputs=[
        gr.inputs.Textbox(label="Question"),
        gr.inputs.Textbox(label="Context")
    ],
    outputs=gr.outputs.Textbox(label="Generated Answer")
)

# Launch the Gradio interface
iface.launch()