Getting Started¶

Prerequisites¶

Please follow instructions in Clara installation to install the docker image and start the container.

Additionally you would like to mount some shared disk/folder for saving all the models, logs and configurations for AIAA server to persist. This will help you to start/stop the container any time without losing the models/configurations for AIAA.

For example:

export NVIDIA_RUNTIME="--runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=0"
export OPTIONS="--shm-size=1g --ulimit memlock=-1 --ulimit stack=67108864"
export SOURCE_DIR=<source dir to store>
export MOUNT_DIR=/aiaa-experiments
export LOCAL_PORT=<the port you want to use>
export REMOTE_PORT=80
export DOCKER_IMAGE="nvcr.io/clara-train-sdk:<version here>"

docker run $NVIDIA_RUNTIME  $OPTIONS -it --rm \
       --ipc=host \
       -p $LOCAL_PORT:$REMOTE_PORT \
       -v $SOURCE_DIR:$MOUNT_DIR \
       $DOCKER_IMAGE \
       /bin/bash

Attention

The system requirements of AIAA depends on how many models you want to load in the server. If your models are big and you don’t have enough system RAM/GPU memory, please load one model at a time.

Attention

If running on TRTIS backend, by default AIAA will put one model instance on every GPU that is visible inside the docker. Please use -e NVIDIA_VISIBLE_DEVICES=<ids of the GPU you want to use> or specify instance group in TRTIS config for fine-grain resource control.

Running AIAA¶

The simple command to start AI-Assisted Annotation server inside the container is: start_aas.sh However, following are the options available while starting AI-Assisted Annotation Server.

Option	Default Value	Example	Description
BASIC
workspace	/var/nvidia/aiaa	`--workspace /aiaa-experiments/aiaa-1`	Workspace Location for AIAA Server to save all the configurations, logs and models. Hence always recommended to to use shared workspace while starting AIAA
debug	false	`--debug true`	Enable Debug level Logging for AIAA Server logs
auto_reload	false	`--auto_reload true`	AIAA Server will auto-reload models whenever the configuration is updated externally. This option should be enabled, when multiple AIAA Server are installed and sharing a common workspace
admin_hosts	*	`--admin_hosts 10.20.1.23,122.34.1.3.4`	Restrict client hosts who can send /admin requests to AIAA server (for e.g. managing models). * (star) will allow any; otherwise provide a list of client IP addresses/prefixes to allow
SSL Support
ssl	false	`--ssl true`	Run AIAA server in ssl mode
ssl_cert_file		`--ssl_cert_file server.cert`	SSL Certificate File Path
ssl_pkey_file		`--ssl_pkey_file server.key`	SSL Key File Path
TF/TRTIS
engine	TRTIS	`--engine AIAA`	Use this option to run AIAA server using Tensorflow (AIAA) or with TRTIS as background inference engine
trtis_ip	localhost	`--trtis_ip 10.18.2.13`	If anything other than 127.0.0.1 or localhost is specified, AIAA will not start a local TRTIS server for inference. Otherwise AIAA will start a local instance of TRTIS server
trtis_port	8001	`--trtis_port 9001`	TRTIS server port (http: 8000, grpc: 8001; monitor: 8002)
trtis_proto	grpc		TRTIS server protocol (http or grpc)
trtis_model_path	<workspace dir>/trtis_models		TRTIS models path
trtis_start_timeout	120		Wait time in seconds for AIAA to make sure TRTIS server is up and running
trtis_model_timeout	30		Wait time in seconds for AIAA to make sure model is loaded and up for serving in TRTIS
FineTune
fine_tune	false	`--fine_tune true`	If set to true, will fine tune the models in AIAA automatically based on samples directory (Model Fine-Tune Setup)
fine_tune_hour	0	`--fine_tune_hour 1`	The scheduled time (0~23) in each day to fine tune all the models
OTHERS
monitor	true	`--monitor true`	If API instrumentation is allowed. If enabled, you can view monitor dashboard by visiting http://host:port/dashboard
client_id	host-80	`--client_id aiaa-instance-1`	If you are deploying AIAA Server on multi node/AWS with shared disk, you might need this to identify each AIAA server accordingly
use_cupy	false	`--use_cupy true`	Use CUPY for performance boost when you have large GPU memory and more than 1 GPU

Note

TRTIS stands for the NVIDIA TensorRT Inference Server, which provides a cloud inferencing solution optimized for NVIDIA GPUs. More information can be found in their github page.

Examples¶

# Run with workspace (AIAA Server will be running at: http://127.0.0.1/)
start_aas.sh --workspace /aiaa-experiments/aiaa-1

# Run AIAA to run TF Inference engine
start_aas.sh --workspace /aiaa-experiments/aiaa-1 --engine AIAA

# Run AIAA to run TRTIS Inference engine (TRTIS will be started at 8001 on localhost)
start_aas.sh --workspace /aiaa-experiments/aiaa-1 --engine TRTIS

# Run AIAA to connect to remote TRTIS Inference engine (Debug only; Some options may not work)
start_aas.sh --workspace /aiaa-experiments/aiaa-1 --engine TRTIS --trtis_ip 10.20.11.34 --trtis_model_path /shared/trtis_models

Stop AIAA¶

The simple command to stop AI-Assisted Annotation server inside the container is: stop_aas.sh

Troubleshooting¶

Workspace¶

AIAA Server uses workspace directory (if not specified default shall be insider docker at the path: /var/nvidia/aiaa) for saving configs, models (Frozen), logs etc. You can shutdown and run the docker multiple times, if you mount an external workspace path while running the docker. See the advanced options while running the docker.

Following are few files/folders explained from workspace directory. Note that INTERNAL means that file or directory is managed by AIAA server and we suggest not to change it.

File/Folder	Description
aiaa_server_config.json	`INTERNAL` - AIAA Server supports automatic loading of models and corresponding configs are saved in this file
aiaa_server_dashboard.db	`INTERNAL` - AIAA Web/API activities are monitored through Flask Dashboard and corresponding data is saved here
downloads/	`INTERNAL` - Temporary downloads from NGC happens here and temporary data is removed after successful import of model into AIAA Server
logs/	`INTERNAL` - AIAA Server logs are stored over here
models/	`INTERNAL` - All serving models in Frozen Format are stored here
trtis_models/	`INTERNAL` - All serving TRTIS models are stored here unless user provides a different trtis_model_path while starting AIAA Server
lib/	`EXTERNAL` - Put your custom transforms/inferences/inference pipelines here and use them in aiaa_config.json (Bring your own models)
samples/{model}/	`EXTERNAL` - Add your train samples here to trigger an incremental training for a model with new samples (Model Fine-Tune Setup)
mmars/{model}/	`INTERNAL/EXTERNAL` - Put your MMAR here to trigger an incremental training for a model with new samples. Also AIAA server archives any imported MMARs here

Note

We suggest you create a new folder to be the workspace so that if you want to delete these files you can just remove that workspace folder.

Logs¶

Once the server is up and running, you can watch or pull the server logs in the browser.

http://127.0.0.1/logs (server will fetch last 100 lines from the current log file)

http://127.0.0.1/logs?lines=-1 to fetch everything from the current log file.

Tip

You don’t have to be in bash mode to see the logs files.

Monitor¶

To track API usage, profiling etc, developers can log in and access monitoring dashboard for the AIAA Server.

http://127.0.0.1/dashboard

Tip

Login Username: admin; Default password: admin

Note

All the URL with 0.0.0.0 or 127.0.0.1 is only accessible if your browser and server are on the same machine. You can use ip a to find your server’s IP address and check it remotely on http://[server_ip]:[port]/logs.

When you run docker, make sure you have docker ports (e.g. http-port: 80) mapped to host machine for external access. You can do that with docker option -p [LOCAL_PORT]:[REMOTE_PORT]. The local port is the host port you want to use while the remote port is the port which AIAA listen to inside the docker container (80 for HTTP, 443 for HTTPS).