EXPERIMENTAL: Download a model and prepare it for building
Synopsis
EXPERIMENTAL: Downloads model files from a provider (e.g. HuggingFace) into a directory and generates a Dockerfile and model.yaml metadata file.
The source can be a HuggingFace model URL/identifier or a local directory path. Use --output-dir to specify where the model files, Dockerfile, and metadata are placed.
After init, use 'nova model build' to build and push the image.
Examples: nova model init meta-llama/Llama-2-7b --output-dir ./llama-2-7b nova model init https://huggingface.co/meta-llama/Llama-2-7b -o ./llama -n my-llama nova model init ./local-model-dir --output-dir ./build
nova model init <source> [flags]
Options
-h, --help help for init
-n, --name string Model name (defaults to name derived from source)
-o, --output-dir string Directory to download model files and generate Dockerfile (default ".")
-p, --provider string Model provider (huggingface, modelscope, mlflow). Auto-detected from URL if not specified.
-v, --version string Model version (default "1.0.0")
Options inherited from parent commands
--debug Write diagnostic output to stderr: HTTP requests and responses, external commands, the resolved instance. Known secrets are redacted, but bodies (app environments, kubeconfigs) can still contain some: don't share the output unreviewed (env: NOVA_DEBUG=1)
--env string (Wandelbots internal) Portal environment for this invocation: prod, stg or dev (default: the environment of the cloud instance, else of the last 'nova login'; env: NOVA_ENV)
--insecure-skip-tls-verify If true, the server's certificate will not be checked for validity (disables TLS verification for all requests)
--instance string Name of the instance to work against for this invocation (overrides the selected instance; env: NOVA_INSTANCE)
SEE ALSO
- nova model - EXPERIMENTAL: Build and push model images