Quick Start Guide

Get up and running with Emotion-LLaMA in under 10 minutes.

Table of Contents

  1. Prerequisites
  2. 5-Minute Setup
    1. Step 1: Clone and Setup Environment
    2. Step 2: Download Pre-trained Models
    3. Step 3: Configure Paths
    4. Step 4: Run the Demo
  3. Try It Out
    1. Using the Web Interface
    2. Example Prompts
  4. Using the API
    1. Python Example
  5. What’s Next?
  6. Need Help?

Prerequisites

Before you begin, ensure you have:

  • ✅ A system with NVIDIA GPU (24GB+ VRAM recommended)
  • ✅ Conda or Miniconda installed
  • ✅ Git installed
  • ✅ At least 50GB free disk space

5-Minute Setup

Step 1: Clone and Setup Environment 

# Clone the repository
git clone https://github.com/ZebangCheng/Emotion-LLaMA.git
cd Emotion-LLaMA

# Create and activate conda environment
conda env create -f environment.yaml
conda activate llama

Step 2: Download Pre-trained Models

Download the following models and place them in the checkpoints/ directory:

  1. Llama-2-7b-chat-hf: Download from Hugging Face
  2. MiniGPT-v2: Download the checkpoint (link in main README)
  3. Emotion-LLaMA Demo Model: Download from Google Drive

Step 3: Configure Paths

Update the model paths in the configuration files:

In minigpt4/configs/models/minigpt_v2.yaml:

llama_model: "checkpoints/Llama-2-7b-chat-hf"

In eval_configs/demo.yaml:

ckpt: "checkpoints/save_checkpoint/Emoation_LLaMA.pth"

Step 4: Run the Demo 

python app.py

Visit http://127.0.0.1:7860 in your browser to try the demo!

Try It Out

Using the Web Interface

  1. Upload a video file (or use one of the example videos from examples/)
  2. Enter a prompt, such as:
    • “What emotion is expressed in this video?”
    • “Describe the facial expressions and tone.”
    • “What is the person feeling and why?”
  3. Click “Submit” to get the model’s response

Example Prompts

For emotion recognition:

[emotion] What is the emotion expressed in this video?

For emotion reasoning:

[reason] What are the facial expressions and vocal tone used in the video? 
What is the intended meaning behind the words? Which emotion does this reflect?

Using the API

You can also use Emotion-LLaMA programmatically via the API.

Python Example 

import json
import requests

url = "http://127.0.0.1:7889/api/predict/"
headers = {"Content-Type": "application/json"}

data = {
    "data": [
        "/path/to/video.mp4",
        "[emotion] What emotion is expressed in this video?"
    ]
}

response = requests.post(url, headers=headers, data=json.dumps(data))
print(response.json())

For more API details, see the API Documentation.

What’s Next?

Now that you have Emotion-LLaMA running, you can:

Need Help?

Happy emotion recognition! 🎉