app.py

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


if torch.cuda.is_available():
    device = torch.device("cuda")
    print("GPU is available!")
else:
    device = torch.device("cpu")
    print("GPU is not available, using CPU.")

# Load the model and config when the script starts
peft_model_id = "phearion/bigbrain-v0.0.1"
config = PeftConfig.from_pretrained(peft_model_id)
model = AutoModelForCausalLM.from_pretrained(
    config.base_model_name_or_path,                          
    torch_dtype=torch.bfloat16)
tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)

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

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, do_sample=True, max_new_tokens=15)

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


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