Torchserve Dashboard using Streamlit
Related blog post
Additional Requirement: torchserve (tested on:v0.5.3, supports: v0.7.0 (no API changes between versions))
Simply run:
pip3 install torchserve-dashboard --user
# torchserve-dashboard [streamlit_options(optional)] -- [config_path(optional)] [model_store(optional)] [log_location(optional)] [metrics_location(optional)]
torchserve-dashboard
#OR change port
torchserve-dashboard --server.port 8105 -- --config_path ./torchserve.properties
#OR provide a custom configuration
torchserve-dashboard -- --config_path ./torchserve.properties --model_store ./model_store
❗ Keep in mind that If you change any of the --config_path
,--model_store
,--metrics_location
,--log_location
options while there is a torchserver already running before starting torch-dashboard they won't come into effect until you stop&start torchserve. These options are used instead of their respective environment variables TS_CONFIG_FILE, METRICS_LOCATION, LOG_LOCATION
.
OR
git clone https://github.com/cceyda/torchserve-dashboard.git
streamlit run torchserve_dashboard/dash.py
#OR
streamlit run torchserve_dashboard/dash.py --server.port 8105 -- --config_path ./torchserve.properties
Example torchserve config:
inference_address=http://127.0.0.1:8443
management_address=http://127.0.0.1:8444
metrics_address=http://127.0.0.1:8445
grpc_inference_port=7070
grpc_management_port=7071
number_of_gpu=0
batch_size=1
model_store=./model_store
If the server doesn't start for some reason check if your ports are already in use!
[15-oct-2020] add scale workers tab
[16-feb-2021] (functionality) make logpath configurable,(functionality)remove model_name requirement,(UI)add cosmetic error messages
[10-may-2021] update config & make it optional. update streamlit. Auto create folders
[31-may-2021] Update to v0.4 (Add workflow API) Refactor out streamlit from api.py.
[30-nov-2021] Update to v0.5, adding support for encrypted model serving (not tested). Update streamlit to v1+
-
Does torchserver keep running in the background?
The torchserver is spawned using
Popen
and keeps running in the background even if you stop the dashboard. -
What about environment variables?
These environment variables are passed to the torchserve command:
ENVIRON_WHITELIST=["LD_LIBRARY_PATH","LC_CTYPE","LC_ALL","PATH","JAVA_HOME","PYTHONPATH","TS_CONFIG_FILE","LOG_LOCATION","METRICS_LOCATION","AWS_ACCESS_KEY_ID", "AWS_SECRET_ACCESS_KEY", "AWS_DEFAULT_REGION"]
-
How to change the logging format of torchserve?
You can set the location of your custom log4j2 config in your configuration file as in here
vmargs=-Dlog4j.configurationFile=file:///path/to/custom/log4j2.xml
-
What is the meaning behind the weird versioning?
The minor follows the compatible torchserve version, patch version reflects the dashboard versioning
Open an issue
- Async?
- Better logging
- Remote only mode