llama-index-multi-modal-llms-huggingface 0.1.0

llama-index multi_modal_llms HuggingFace integration by [Cihan Yalçın](https://www.linkedin.com/in/chanyalcin/)

LlamaIndex Multi_Modal_Llms Integration: Huggingface

This project integrates Hugging Face's multimodal language models into the LlamaIndex framework, enabling advanced multimodal capabilities for various AI applications.

Features

• Seamless integration of Hugging Face multimodal models with LlamaIndex
• Support for multiple state-of-the-art vision-language models and their finetunes:
  • Qwen2 Vision
  • Florence2
  • Phi-3.5 Vision
  • PaLI-Gemma
• Easy-to-use interface for multimodal tasks like image captioning and visual question answering
• Configurable model parameters for fine-tuned performance

---

Author of that Integration GitHub | LinkedIn | Email

Installation

pip install llama-index-multi-modal-llms-huggingface

Make sure to set your Hugging Face API token as an environment variable:

export HF_TOKEN=your_huggingface_token_here

Usage

Here's a basic example of how to use the Hugging Face multimodal integration:

from llama_index.multi_modal_llms.huggingface import HuggingFaceMultiModal
from llama_index.schema import ImageDocument

# Initialize the model
model = HuggingFaceMultiModal.from_model_name("Qwen/Qwen2-VL-2B-Instruct")

# Prepare your image and prompt
image_document = ImageDocument(image_path="path/to/your/image.jpg")
prompt = "Describe this image in detail."

# Generate a response
response = model.complete(prompt, image_documents=[image_document])

print(response.text)

You can also refer to this Colab notebook

Supported Models

1. Qwen2VisionMultiModal
2. Florence2MultiModal
3. Phi35VisionMultiModal
4. PaliGemmaMultiModal

Each model has its unique capabilities and can be selected based on your specific use case.

Configuration

You can configure various parameters when initializing a model:

model = HuggingFaceMultiModal(
    model_name="Qwen/Qwen2-VL-2B-Instruct",
    device="cuda",  # or "cpu"
    torch_dtype=torch.float16,
    max_new_tokens=100,
    temperature=0.7,
)

Limitations

• Async streaming is not supported for any of the models.
• Some models have specific requirements or limitations. Please refer to the individual model classes for details.

---

Author of that Integration GitHub | LinkedIn | Email