MultimodalQnA Application

Suppose you possess a set of videos and wish to perform question-answering to extract insights from these videos. To respond to your questions, it typically necessitates comprehension of visual cues within the videos, knowledge derived from the audio content, or often a mix of both these visual elements and auditory facts. The MultimodalQnA framework offers an optimal solution for this purpose.

MultimodalQnA addresses your questions by dynamically fetching the most pertinent multimodal information (frames, transcripts, and/or captions) from your collection of videos. For this purpose, MultimodalQnA utilizes BridgeTower model, a multimodal encoding transformer model which merges visual and textual data into a unified semantic space. During the video ingestion phase, the BridgeTower model embeds both visual cues and auditory facts as texts, and those embeddings are then stored in a vector database. When it comes to answering a question, the MultimodalQnA will fetch its most relevant multimodal content from the vector store and feed it into a downstream Large Vision-Language Model (LVM) as input context to generate a response for the user.

The MultimodalQnA architecture shows below:

MultimodalQnA is implemented on top of GenAIComps, the MultimodalQnA Flow Chart shows below:

flowchart LR %% Colors %% classDef blue fill:#ADD8E6,stroke:#ADD8E6,stroke-width:2px,fill-opacity:0.5 classDef orange fill:#FBAA60,stroke:#ADD8E6,stroke-width:2px,fill-opacity:0.5 classDef orchid fill:#C26DBC,stroke:#ADD8E6,stroke-width:2px,fill-opacity:0.5 classDef invisible fill:transparent,stroke:transparent; style MultimodalQnA-MegaService stroke:#000000 %% Subgraphs %% subgraph MultimodalQnA-MegaService["MultimodalQnA-MegaService"] direction LR EM([Embedding <br>]):::blue RET([Retrieval <br>]):::blue LVM([LVM <br>]):::blue end subgraph User Interface direction TB a([User Input Query]):::orchid Ingest([Ingest data]):::orchid UI([UI server<br>]):::orchid end subgraph MultimodalQnA GateWay direction LR invisible1[ ]:::invisible GW([MultimodalQnA GateWay<br>]):::orange end subgraph . X([OPEA Microservice]):::blue Y{{Open Source Service}} Z([OPEA Gateway]):::orange Z1([UI]):::orchid end TEI_EM{{Embedding service <br>}} VDB{{Vector DB<br><br>}} R_RET{{Retriever service <br>}} DP([Data Preparation<br>]):::blue LVM_gen{{LVM Service <br>}} %% Data Preparation flow %% Ingest data flow direction LR Ingest[Ingest data] -->|a| UI UI -->|b| DP DP <-.->|c| TEI_EM %% Questions interaction direction LR a[User Input Query] -->|1| UI UI -->|2| GW GW <==>|3| MultimodalQnA-MegaService EM ==>|4| RET RET ==>|5| LVM %% Embedding service flow direction TB EM <-.->|3'| TEI_EM RET <-.->|4'| R_RET LVM <-.->|5'| LVM_gen direction TB %% Vector DB interaction R_RET <-.->|d|VDB DP <-.->|e|VDB

This MultimodalQnA use case performs Multimodal-RAG using LangChain, Redis VectorDB and Text Generation Inference on Intel Gaudi2 and Intel Xeon Scalable Processors, and we invite contributions from other hardware vendors to expand the example.

The Intel Gaudi2 accelerator supports both training and inference for deep learning models in particular for LLMs. Visit Habana AI products for more details.

In the below, we provide a table that describes for each microservice component in the MultimodalQnA architecture, the default configuration of the open source project, hardware, port, and endpoint.

Gaudi default compose.yaml

MicroService	Open Source Project	HW	Port	Endpoint
Embedding	Langchain	Xeon	6000	/v1/embeddings
Retriever	Langchain, Redis	Xeon	7000	/v1/multimodal_retrieval
LVM	Langchain, TGI	Gaudi	9399	/v1/lvm
Dataprep	Redis, Langchain, TGI	Gaudi	6007	/v1/generate_transcripts, /v1/generate_captions

Required Models

By default, the embedding and LVM models are set to a default value as listed below:

Service	Model
embedding-multimodal	BridgeTower/bridgetower-large-itm-mlm-gaudi
LVM	llava-hf/llava-v1.6-vicuna-13b-hf

You can choose other LVM models, such as llava-hf/llava-1.5-7b-hf and llava-hf/llava-1.5-13b-hf, as needed.

Deploy MultimodalQnA Service

The MultimodalQnA service can be effortlessly deployed on either Intel Gaudi2 or Intel XEON Scalable Processors.

Currently we support deploying MultimodalQnA services with docker compose.

Setup Environment Variable

To set up environment variables for deploying MultimodalQnA services, follow these steps:

1. Set the required environment variables:

   # Example: export host_ip=$(hostname -I | awk '{print $1}')
   export host_ip="External_Public_IP"
   # Example: no_proxy="localhost, 127.0.0.1, 192.168.1.1"
   export no_proxy="Your_No_Proxy"
   

2. If you are in a proxy environment, also set the proxy-related environment variables:

   export http_proxy="Your_HTTP_Proxy"
   export https_proxy="Your_HTTPs_Proxy"
   

3. Set up other environment variables:

   Notice that you can only choose one command below to set up envs according to your hardware. Other that the port numbers may be set incorrectly.

   # on Gaudi
   source ./docker_compose/intel/hpu/gaudi/set_env.sh
   # on Xeon
   source ./docker_compose/intel/cpu/xeon/set_env.sh
   

Deploy MultimodalQnA on Gaudi

Refer to the Gaudi Guide to build docker images from source.

Find the corresponding compose.yaml.

cd GenAIExamples/MultimodalQnA/docker_compose/intel/hpu/gaudi/
docker compose -f compose.yaml up -d

Notice: Currently only the Habana Driver 1.17.x is supported for Gaudi.

Deploy MultimodalQnA on Xeon

Refer to the Xeon Guide for more instructions on building docker images from source.

Find the corresponding compose.yaml.

cd GenAIExamples/MultimodalQnA/docker_compose/intel/cpu/xeon/
docker compose -f compose.yaml up -d

MultimodalQnA Demo on Gaudi2