app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline

# Replace with your model name
#MODEL_NAME = "ssirikon/Gemma7b-bnb-Unsloth"
#MODEL_NAME = "unsloth/gemma-7b-bnb-4bit"
MODEL_NAME = "unsloth/mistral-7b-bnb-4bit"

# Load the model and tokenizer
model = AutoModelForCausalLM.from_pretrained(
    MODEL_NAME, 
    device_map="auto", 
    torch_dtype=torch.float16,
    load_in_4bit=True,  # Load the model in 4-bit precision
    # Removed the unsupported argument
)
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)

# **Change 1: Set `llm_int8_skip_modules` to avoid deep copy**
#model.quantization_config.llm_int8_skip_modules = ['lm_head'] 

# Create a pipeline for text generation
generator = pipeline(
    task="summarization",
    model=model,
    tokenizer=tokenizer,
    max_new_tokens=50,  # Adjust as needed
    do_sample=True,
    top_k=10,
    num_return_sequences=1,
    eos_token_id=tokenizer.eos_token_id,
)

def generate_text(email):
    result = generator("Generate a subject line for the following email.\n"+email)
    return result[0]["generated_text"]


# Create a Gradio interface
demo = gr.Interface(
    fn=generate_text,
    inputs=gr.Textbox(lines=5, label="Enter your Email here:"),
    outputs=gr.Textbox(label="Generated Subject"),
    title="Email Subject Generation demo",
    description="Enter an email and let the model generate the subject for you!",
)

demo.launch(debug=True)