app.py

import gradio as gr
from peft import PeftModel, PeftConfig
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# Load the model and config when the script starts
config = PeftConfig.from_pretrained("phearion/bigbrain-v0.0.1")
model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-v0.1")
model = PeftModel.from_pretrained(model, "phearion/bigbrain-v0.0.1")

# Move the model to the device
model = model.to(device)


# Load the tokenizer
tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1")



def greet(text):
    batch = tokenizer(f"'{text}' ->: ", return_tensors='pt')
    
    # Use torch.no_grad to disable gradient calculation
    with torch.no_grad():
        output_tokens = model.generate(**batch, max_new_tokens=20)

    return tokenizer.decode(output_tokens[0], skip_special_tokens=True)


iface = gr.Interface(fn=greet, inputs="text", outputs="text")
iface.launch()