Build MegaService of Document Summarization on Gaudi

This document outlines the deployment process for a Document Summarization application utilizing the GenAIComps microservice pipeline on Intel Gaudi server. The steps include Docker image creation, container deployment via Docker Compose, and service execution to integrate microservices such as llm. We will publish the Docker images to Docker Hub soon, which will simplify the deployment process for this service.

🚀 Build Docker Images

1. Build MicroService Docker Image

First of all, you need to build Docker Images locally and install the python package of it.

git clone https://github.com/opea-project/GenAIComps.git
cd GenAIComps

Audio to text Service

The Audio to text Service is another service for converting audio to text. Follow these steps to build and run the service:

docker build -t opea/dataprep-audio2text:latest --build-arg https_proxy=$https_proxy --build-arg http_proxy=$http_proxy -f comps/dataprep/multimedia2text/audio2text/Dockerfile .

Video to Audio Service

The Video to Audio Service extracts audio from video files. Follow these steps to build and run the service:

docker build -t opea/dataprep-video2audio:latest --build-arg https_proxy=$https_proxy --build-arg http_proxy=$http_proxy -f comps/dataprep/multimedia2text/video2audio/Dockerfile .

Multimedia to Text Service

The Multimedia to Text Service transforms multimedia data to text data. Follow these steps to build and run the service:

docker build -t opea/dataprep-multimedia2text:latest --build-arg https_proxy=$https_proxy --build-arg http_proxy=$http_proxy -f comps/dataprep/multimedia2text/Dockerfile .

2. Build MegaService Docker Image

To construct the Mega Service, we utilize the GenAIComps microservice pipeline within the docsum.py Python script. Build the MegaService Docker image via below command:

git clone https://github.com/opea-project/GenAIExamples
cd GenAIExamples/DocSum/
docker build -t opea/docsum:latest --build-arg https_proxy=$https_proxy --build-arg http_proxy=$http_proxy -f Dockerfile .

3. Build UI Docker Image

Several UI options are provided. If you need to work with multimedia documents, .doc, or .pdf files, suggested to use Gradio UI.

Gradio UI

Build the Gradio UI frontend Docker image using the following command:

cd GenAIExamples/DocSum/ui
docker build -t opea/docsum-gradio-ui:latest --build-arg https_proxy=$https_proxy --build-arg http_proxy=$http_proxy -f docker/Dockerfile.gradio .

Svelte UI

Build the frontend Docker image via below command:

cd GenAIExamples/DocSum/ui
docker build -t opea/docsum-ui:latest --build-arg https_proxy=$https_proxy --build-arg http_proxy=$http_proxy -f docker/Dockerfile .

React UI

Build the frontend Docker image via below command:

cd GenAIExamples/DocSum/ui
export BACKEND_SERVICE_ENDPOINT="http://${host_ip}:8888/v1/docsum"
docker build -t opea/docsum-react-ui:latest --build-arg BACKEND_SERVICE_ENDPOINT=$BACKEND_SERVICE_ENDPOINT -f ./docker/Dockerfile.react .

docker build -t opea/docsum-react-ui:latest --build-arg BACKEND_SERVICE_ENDPOINT=$BACKEND_SERVICE_ENDPOINT --build-arg https_proxy=$https_proxy --build-arg http_proxy=$http_proxy  -f ./docker/Dockerfile.react .

🚀 Start Microservices and MegaService

Required Models

Default model is "Intel/neural-chat-7b-v3-3". Change "LLM_MODEL_ID" environment variable in commands below if you want to use another model.

export LLM_MODEL_ID="Intel/neural-chat-7b-v3-3"

When using gated models, you also need to provide HuggingFace token to "HUGGINGFACEHUB_API_TOKEN" environment variable.

Setup Environment Variable

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

Set the required environment variables:

# Example: host_ip="192.168.1.1"
export host_ip="External_Public_IP"
# Example: no_proxy="localhost, 127.0.0.1, 192.168.1.1"
export no_proxy="Your_No_Proxy"
export HUGGINGFACEHUB_API_TOKEN="Your_Huggingface_API_Token"

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"
```

Set up other environment variables:

source GenAIExamples/DocSum/docker_compose/set_env.sh

Start Microservice Docker Containers

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

You will have the following Docker Images:

opea/docsum-ui:latest
opea/docsum:latest
opea/llm-docsum-tgi:latest
opea/whisper:latest
opea/dataprep-audio2text:latest
opea/dataprep-multimedia2text:latest
opea/dataprep-video2audio:latest

Validate Microservices

TGI Service

curl http://${host_ip}:8008/generate \
  -X POST \
  -d '{"inputs":"What is Deep Learning?","parameters":{"max_new_tokens":17, "do_sample": true}}' \
  -H 'Content-Type: application/json'

LLM Microservice

curl http://${host_ip}:9000/v1/chat/docsum \
  -X POST \
  -d '{"query":"Text Embeddings Inference (TEI) is a toolkit for deploying and serving open source text embeddings and sequence classification models. TEI enables high-performance extraction for the most popular models, including FlagEmbedding, Ember, GTE and E5."}' \
  -H 'Content-Type: application/json'

Whisper Microservice

 curl http://${host_ip}:7066/v1/asr \
     -X POST \
     -d '{"audio":"UklGRigAAABXQVZFZm10IBIAAAABAAEARKwAAIhYAQACABAAAABkYXRhAgAAAAEA"}' \
     -H 'Content-Type: application/json'

Expected output:

  {"asr_result":"you"}

Audio2Text Microservice

 curl http://${host_ip}:9199/v1/audio/transcriptions \
     -X POST \
     -d '{"byte_str":"UklGRigAAABXQVZFZm10IBIAAAABAAEARKwAAIhYAQACABAAAABkYXRhAgAAAAEA"}' \
     -H 'Content-Type: application/json'

Expected output:

  {"downstream_black_list":[],"id":"--> this will be different id number for each run <--","query":"you"}

Multimedia to text Microservice

 curl http://${host_ip}:7079/v1/multimedia2text \
     -X POST \
     -d '{"audio":"UklGRigAAABXQVZFZm10IBIAAAABAAEARKwAAIhYAQACABAAAABkYXRhAgAAAAEA"}' \
     -H 'Content-Type: application/json'

Expected output:

  {"downstream_black_list":[],"id":"--> this will be different id number for each run <--","query":"you"}

MegaService

Text:

## json input
curl -X POST http://${host_ip}:8888/v1/docsum \
     -H "Content-Type: application/json" \
     -d '{"type": "text", "messages": "Text Embeddings Inference (TEI) is a toolkit for deploying and serving open source text embeddings and sequence classification models. TEI enables high-performance extraction for the most popular models, including FlagEmbedding, Ember, GTE and E5."}'

# form input. Use English mode (default).
curl http://${host_ip}:8888/v1/docsum \
    -H "Content-Type: multipart/form-data" \
    -F "type=text" \
    -F "messages=Text Embeddings Inference (TEI) is a toolkit for deploying and serving open source text embeddings and sequence classification models. TEI enables high-performance extraction for the most popular models, including FlagEmbedding, Ember, GTE and E5." \
    -F "max_tokens=32" \
    -F "language=en" \
    -F "stream=True"

# Use Chinese mode.
curl http://${host_ip}:8888/v1/docsum \
    -H "Content-Type: multipart/form-data" \
    -F "type=text" \
    -F "messages=2024年9月26日，北京——今日，英特尔正式发布英特尔® 至强® 6性能核处理器（代号Granite Rapids），为AI、数据分析、科学计算等计算密集型业务提供卓越性能。" \
    -F "max_tokens=32" \
    -F "language=zh" \
    -F "stream=True"

# Upload file
curl http://${host_ip}:8888/v1/docsum \
   -H "Content-Type: multipart/form-data" \
   -F "type=text" \
   -F "messages=" \
   -F "files=@/path to your file (.txt, .docx, .pdf)" \
   -F "max_tokens=32" \
   -F "language=en" \

Audio and Video file uploads are not supported in docsum with curl request, please use the Gradio-UI.

Audio:

curl -X POST http://${host_ip}:8888/v1/docsum \
   -H "Content-Type: application/json" \
   -d '{"type": "audio", "messages": "UklGRigAAABXQVZFZm10IBIAAAABAAEARKwAAIhYAQACABAAAABkYXRhAgAAAAEA"}'

curl http://${host_ip}:8888/v1/docsum \
   -H "Content-Type: multipart/form-data" \
   -F "type=audio" \
   -F "messages=UklGRigAAABXQVZFZm10IBIAAAABAAEARKwAAIhYAQACABAAAABkYXRhAgAAAAEA" \
   -F "max_tokens=32" \
   -F "language=en" \
   -F "stream=True"

Video:

curl -X POST http://${host_ip}:8888/v1/docsum \
   -H "Content-Type: application/json" \
   -d '{"type": "video", "messages": "convert your video to base64 data type"}'

curl http://${host_ip}:8888/v1/docsum \
   -H "Content-Type: multipart/form-data" \
   -F "type=video" \
   -F "messages=convert your video to base64 data type" \
   -F "max_tokens=32" \
   -F "language=en" \
   -F "stream=True"

MegaService with long context

If you want to deal with long context, can set following parameters and select suitable summary type.
- "summary_type": can be "auto", "stuff", "truncate", "map_reduce", "refine", default is "auto"
- "chunk_size": max token length for each chunk. Set to be different default value according to "summary_type".
- "chunk_overlap": overlap token length between each chunk, default is 0.1*chunk_size
summary_type=auto

"summary_type" is set to be "auto" by default, in this mode we will check input token length, if it exceed MAX_INPUT_TOKENS, summary_type will automatically be set to refine mode, otherwise will be set to stuff mode.
```
curl http://${host_ip}:8888/v1/docsum \
   -H "Content-Type: multipart/form-data" \
   -F "type=text" \
   -F "messages=" \
   -F "max_tokens=32" \
   -F "files=@/path to your file (.txt, .docx, .pdf)" \
   -F "language=en" \
   -F "summary_type=auto"
```
summary_type=stuff

In this mode LLM generate summary based on complete input text. In this case please carefully set MAX_INPUT_TOKENS and MAX_TOTAL_TOKENS according to your model and device memory, otherwise it may exceed LLM context limit and raise error when meet long context.
```
curl http://${host_ip}:8888/v1/docsum \
   -H "Content-Type: multipart/form-data" \
   -F "type=text" \
   -F "messages=" \
   -F "max_tokens=32" \
   -F "files=@/path to your file (.txt, .docx, .pdf)" \
   -F "language=en" \
   -F "summary_type=stuff"
```
summary_type=truncate

Truncate mode will truncate the input text and keep only the first chunk, whose length is equal to min(MAX_TOTAL_TOKENS - input.max_tokens - 50, MAX_INPUT_TOKENS)
```
curl http://${host_ip}:8888/v1/docsum \
   -H "Content-Type: multipart/form-data" \
   -F "type=text" \
   -F "messages=" \
   -F "max_tokens=32" \
   -F "files=@/path to your file (.txt, .docx, .pdf)" \
   -F "language=en" \
   -F "summary_type=truncate"
```
summary_type=map_reduce

Map_reduce mode will split the inputs into multiple chunks, map each document to an individual summary, then consolidate those summaries into a single global summary. streaming=True is not allowed here.

In this mode, default chunk_size is set to be min(MAX_TOTAL_TOKENS - input.max_tokens - 50, MAX_INPUT_TOKENS)
```
curl http://${host_ip}:8888/v1/docsum \
   -H "Content-Type: multipart/form-data" \
   -F "type=text" \
   -F "messages=" \
   -F "max_tokens=32" \
   -F "files=@/path to your file (.txt, .docx, .pdf)" \
   -F "language=en" \
   -F "summary_type=map_reduce"
```
summary_type=refine

Refin mode will split the inputs into multiple chunks, generate summary for the first one, then combine with the second, loops over every remaining chunks to get the final summary.

In this mode, default chunk_size is set to be min(MAX_TOTAL_TOKENS - 2 * input.max_tokens - 128, MAX_INPUT_TOKENS).
```
curl http://${host_ip}:8888/v1/docsum \
   -H "Content-Type: multipart/form-data" \
   -F "type=text" \
   -F "messages=" \
   -F "max_tokens=32" \
   -F "files=@/path to your file (.txt, .docx, .pdf)" \
   -F "language=en" \
   -F "summary_type=refine"
```

More detailed tests can be found here cd GenAIExamples/DocSum/test

🚀 Launch the UI

Several UI options are provided. If you need to work with multimedia documents, .doc, or .pdf files, suggested to use Gradio UI.

Gradio UI

Open this URL http://{host_ip}:5173 in your browser to access the Gradio based frontend.

🚀 Launch the Svelte UI

Open this URL http://{host_ip}:5173 in your browser to access the Svelte based frontend.

Here is an example for summarizing a article.

🚀 Launch the React UI (Optional)

To access the React-based frontend, modify the UI service in the compose.yaml file. Replace docsum-xeon-ui-server service with the docsum-xeon-react-ui-server service as per the config below:

docsum-gaudi-react-ui-server:
  image: ${REGISTRY:-opea}/docsum-react-ui:${TAG:-latest}
  container_name: docsum-gaudi-react-ui-server
  depends_on:
    - docsum-gaudi-backend-server
  ports:
    - "5174:80"
  environment:
    - no_proxy=${no_proxy}
    - https_proxy=${https_proxy}
    - http_proxy=${http_proxy}
    - DOC_BASE_URL=${BACKEND_SERVICE_ENDPOINT}

Open this URL http://{host_ip}:5175 in your browser to access the frontend.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

README.md

README.md

Build MegaService of Document Summarization on Gaudi

🚀 Build Docker Images

1. Build MicroService Docker Image

Audio to text Service

Video to Audio Service

Multimedia to Text Service

2. Build MegaService Docker Image

3. Build UI Docker Image

Gradio UI

Svelte UI

React UI

🚀 Start Microservices and MegaService

Required Models

Setup Environment Variable

Start Microservice Docker Containers

Validate Microservices

🚀 Launch the UI

Gradio UI

🚀 Launch the Svelte UI

🚀 Launch the React UI (Optional)

Files

README.md

Latest commit

History

README.md

File metadata and controls

Build MegaService of Document Summarization on Gaudi

🚀 Build Docker Images

1. Build MicroService Docker Image

Audio to text Service

Video to Audio Service

Multimedia to Text Service

2. Build MegaService Docker Image

3. Build UI Docker Image

Gradio UI

Svelte UI

React UI

🚀 Start Microservices and MegaService

Required Models

Setup Environment Variable

Start Microservice Docker Containers

Validate Microservices

🚀 Launch the UI

Gradio UI

🚀 Launch the Svelte UI

🚀 Launch the React UI (Optional)