app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch

# Load model and tokenizer once on startup
tokenizer = AutoTokenizer.from_pretrained("Salesforce/codet5p-220m")
model = AutoModelForSeq2SeqLM.from_pretrained("Salesforce/codet5p-220m")
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = model.to(device)

def generate(prompt):
    inputs = tokenizer(prompt, return_tensors="pt").to(device)
    outputs = model.generate(
        **inputs,
        max_length=2048,
        num_beams=3,
        early_stopping=True,
        no_repeat_ngram_size=3,
        eos_token_id=tokenizer.eos_token_id,
        pad_token_id=tokenizer.pad_token_id,
    )
    output_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return output_text

# Create Gradio interface
iface = gr.Interface(
    fn=generate,
    inputs=gr.Textbox(lines=10, label="Input Prompt"),
    outputs=gr.Textbox(label="Generated Output"),
    title="LLaMA 7B Server",
    description="A web interface for interacting with the LLaMA 7B model.",
    allow_flagging="never"
)

# Launch the interface
if __name__ == "__main__":
    iface.launch(server_name="0.0.0.0", server_port=7860)