This demo includes a container for runing Lag Llama in Snowflake Container Services. The container includes the Lag Llama model and the GluonTS packages necessary for getting predictions. A sample notebook is included, but you will have to connect to your own time series data sources (a sample dataset is provided in the repo).
-
Build the Docker image
docker build --rm --platform linux/amd64 -t lagllama2 .
-
Tag and push it to Snowpark container services image repo docker tag lagllama2 Based on the location of YOUR image folder: sfsenorthamerica-polaris2.registry.snowflakecomputing.com/mistral_vllm_db/public/images/lagllama2
docker push sfsenorthamerica-polaris2.registry.snowflakecomputing.com/mistral_vllm_db/public/images/lagllama2
-
Push the lagllama_spec.yaml to YOUR stage. Make sure the spec file is set to your environment.
-
Create the service and navigate to the endpoint to access the jupyter environment.
- Start in jupyter server
- Navigate one level down to ll_github folder
- Open the LagLlama.ipynb folder
- To run the demo model, upload the data, timedata.csv which is in the github here
- The notebook will show you the data, run predictions, plot the predictions, and provide evaluation statistics