Deploying Your Custom Model into Jarvis

This section provides a brief overview on the two main tools used in the deploy process:

1. The build phase using jarvis-build.

2. The deploy phase using jarvis-deploy.

Build process

For your custom trained model, refer to the corresponding section (ASR, NLP, TTS) for your model type for the jarvis-build phase. At the end of this phase, you would have the Jarvis Model Intermediate Representation (JMIR) archive for your custom model.

Deploy process

At this point, you already have your Jarvis Model Intermediate Representation (JMIR) archive. Now, you have two options for deploying this JMIR.

Option 1: Use the Quick Start scripts (jarvis_init.sh and jarvis_start.sh) with the appropriate parameters in config.sh,

Option 2: Manually run jarvis-deploy and then start jarvis-server with the target model repo.

Option 1: Using Quick Start Scripts to Deploy Your Models (Recommended path)

The Quick Start scripts (jarvis_init.sh and jarvis_start.sh) uses a particular directory for its operations. This directory is defined by the variable $jarvis_model_loc specified in config.sh. By default this is set to use a Docker volume, however, you can specify any local directory to this variable.

By default, the jarvis_init.sh Quick Start script performs the following:

• Downloads the JMIRs defined and enabled in config.sh, from NGC, into a subdirectory at $jarvis_model_loc, specifically $jarvis_model_loc/jmir.

• Executes jarvis-deploy for each of the JMIRs at $jarvis_model_loc/jmir to generate their corresponding Triton Inference Server model repository at $jarvis_model_loc/models.

When you execute jarvis_start.sh, it starts with the jarvis-speech container by mounting this $jarvis_model_loc directory to /data inside the container.

To deploy your own custom JMIR, or set of JMIRs, you would simply need to place them inside the $jarvis_model_loc/jmir directory. Ensure that you have defined a directory (that you have access to) in the $jarvis_model_loc variable in config.sh, since you will need to copy over your JMIRs in its subdirectory. If the subdirectory $jarvis_model_loc/jmir does not exist, then you’d need to create it and then copy your custom JMIRs there.

If you would like to skip the downloading of default JMIRs from NGC, then you can set the variable $use_existing_jmirs to true. After your custom JMIRs are inside this $jarvis_model_loc/jmir directory, you can run jarvis_init.sh which will execute jarvis-deploy on your custom JMIRs along with any other JMIRs that are present on that directory and generate the Triton Inferece Server model repo at $jarvis_model_loc/models.

Next, you can run jarvis_start.sh and it will start the jarvis-speech container and load your custom models along with any other models that are present at $jarvis_model_loc/models. If you only want to load your specific models, ensure that $jarvis_model_loc/models is empty or the /models directory is not present before you run jarvis_init.sh. The script jarvis_init.sh will create the subdirectories /jmir and /models if they are not already there.

For more information about seeing logs and using client containers for testing your models, refer to the Server Deployment > Local (Docker) section.

Option 2: Using jarvis-deploy and the Jarvis Speech Container (Advanced)

1. Execute jarvis-deploy. Refer to the Deploy section in Services and Models > Overview for a brief overview on jarvis-deploy.

   The above command will create the Triton Inference Server model repository at /data/models. If you want to write to any other location other than /data/models, this will require some more manual changes in the embedded artifact directories within the configs within some of the Triton Inference Server model repositories that has model specific artifacts such as class labels. Therefore, stick with /data/models unless you are familiar with Triton Inference Server Model repository configurations.

2. Manually start the jarvis-server Docker container using docker run.

   After the Triton Inference Server model repository for your custom model is generated, start the Jarvis server on that target repo. The following command assumes you generated the model repo at /data/models.

   docker run -d --gpus 1 --init --shm-size=1G --ulimit memlock=-1 --ulimit stack=67108864 \
           -v /data:/data             \
           -p 50051                            \
           -e '\''CUDA_VISIBLE_DEVICES=0'\''   \
           --name jarvis-speech                \
           jarvis-api                          \
           start-jarvis  --jarvis-uri=0.0.0.0:50051 --nlp_service=true --asr_service=true --tts_service=true
   

   This will launch the Jarvis Speech Service API server similar to the Quick Start script jarvis_start.sh.

   Example output:

   Starting Jarvis Speech Services
   > Waiting for Triton server to load all models...retrying in 10 seconds
   > Waiting for Triton server to load all models...retrying in 10 seconds
   > Waiting for Triton server to load all models...retrying in 10 seconds
   > Triton server is ready…
   

3. Verify that the servers have started correctly and check that the output of docker logs jarvis-speech shows:

   I0428 03:14:50.440943 1 jarvis_server.cc:66] TTS Server connected to Triton Inference Server at 0.0.0.0:8001
   I0428 03:14:50.440943 1 jarvis_server.cc:66] NLP Server connected to Triton Inference Server at 0.0.0.0:8001
   I0428 03:14:50.440951 1 jarvis_server.cc:68] ASR Server connected to Triton Inference Server at 0.0.0.0:8001
   I0428 03:14:50.440955 1 jarvis_server.cc:71] Jarvis Conversational AI Server listening on 0.0.0.0:50051