Version: 1.66

v1.49

info

Make sure your installation is updated to at least Deeploy v1.42.0 before installing v1.49.0

Release notes

Refactored endpoints that returned a response in the format of { data: { ... }, message: "" }. Instead, the content from the data key is directly returned. Full list of affected endpoints
Dropped support for the old commit and sort query parameter on the /workspaces/:workspaceId/deployments/:deploymentId/requestLogs endpoint. Instead of commit=example, sortBy=createdAt or direction=desc, please use the new RHS param (e.g. commit=in:exampleOne,exampleTwo, sort:createdAt:desc)

Updated to KServe 0.15 with relevant CRDs
Bumped up framework versions: XGBoost 2.1.1, Sklearn 1.5.1, LGBM 4.5.0. For the full list consult our supported versions. Added LLM endpoints for generative Hugging Face models (KServe Deployments and external Deployments): /chat/completions, /completions. Added an embeddings endpoint for embedding Hugging Face models (KServe Deployments and external Deployments): /embeddings.

info

Transformers on the /completions, /chat/completions and /embeddings endpoints are not supported for KServe Deployments
Generative and embedding Hugging Face models do not support /predict and /explain

Objects in the audit log details are now stringified
Added audit logs for imported documentation and Deployment compliance templates
Fetch the raw request and response body with the includeRawBody query param on the /workspaces/:workspace/deployments/:deploymentId/requestLogs endpoint

Fixed an issue where the histogram of prediction data was calculated incorrectly in some cases
Show correct popups on the Deployment test page when saving log and when erroring
Fixed an issue with saving 'Docker credentials created' audit logs
Fixed an issue where the documentation completion progress bar would occasionally display an incorrect value
Fixed a scroll issue where upon navigation the scroll view does not move to the top
Fixed an overlapping component issue with the cookie banner
Fixed an issue where there was no model name when deploying Hugging Face model without standard explainer