app.py

import gradio as gr
import torch
from transformers import AutoModel, AutoTokenizer, AutoModelForCausalLM
import deepspeed

# Model name
model_name = "OpenGVLab/InternVideo2_5_Chat_8B"

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)

# Load model efficiently
model = AutoModelForCausalLM.from_pretrained(
    model_name, 
    trust_remote_code=True,
    torch_dtype=torch.float16,  # Use float16 for lower memory usage
    device_map="auto", # Automatically place model on available GPU
    deepspeed={"stage": 3}      # Enable DeepSpeed ZeRO-3
    
)

# Define inference function
def chat_with_model(prompt):
    inputs = tokenizer(prompt, return_tensors="pt").to("cuda")  # Move inputs to GPU
    output = model.generate(**inputs, max_length=200)
    return tokenizer.decode(output[0], skip_special_tokens=True)

# Create Gradio UI
demo = gr.Interface(
    fn=chat_with_model,
    inputs=gr.Textbox(placeholder="Type your prompt here..."),
    outputs="text",
    title="InternVideo2.5 Chatbot",
    description="A chatbot powered by InternVideo2_5_Chat_8B.",
    theme="compact"
)

# Run the Gradio app
if __name__ == "__main__":
    demo.launch()