21st August 2020: New release details
More articles in Release Notes & Updates
1. Support for Zoom in/out features in Builder
- User can zoom in/out by clicking on cmd+scroll for Mac and ctrl+scroll for Windows
- User can apply “fit to canvas” option to fit the whole dag in the same view
- User can apply “Actual Size option” to bring the dag back to the actual size
2. Library Manager support.
- User added Library and System libraries will be shown in Library manager page in Administration section
- User can search and sort both User and System Libraries
- User can add/update/delete libraries in the User library section
3. Layer shapes are shown on the layer itself instead of on the connection
4. Support for changing Pipeline variables and Resource requirements for Blocks while Pipeline Run
- User can change Pipeline Variables on click of run icon from pipeline builder
- User can change Resource Requirements for Blocks on click of run icon from pipeline builder
5. Support for adding Pipeline Variables from the Block
- Users can add Pipeline Variables from Blocks as well, by clicking on the add variable which is shown on the click of “var” icon.
6. Support for Pipeline Rerun from Pipeline Runs page
- On click of rerun from Pipeline Runs page, users cannot change pipeline variables and resource requirements for Blocks. Users can only choose the engine to run on.
7. Support for creating Duplicate Layers and Share Weights for Groups in Model Builder
- User can right-click on Layer/Group and create a duplicate of that layer
- User can right click on a block and enable sharing weights between Groups. When Groups are sharing weights, both Groups will have identical content. If change is made in one of the shared groups, the same changes will be replicated in other Groups which are sharing weights with the Group.
8. Users can provision GPU templates.
9. Support for Custom Loss
- Export model from SDK to Model Builder
- Import model from Model Builder to SDK
10. Support for Custom Metric (same as Custom Loss)
11. Model components
- Conv2d Depthwise
- Conv2d LSTM
- Conv3d Transpose
- SSIM Metric
12. Estimation of memory consumption of DL blocks before running
- It’s the estimated Memory required for the Train block which will be calculated based on the model selected and given batch size and will be updated in resource requirements of that block.
13. Integration of widget with engine pipeline runs
14. Set of ML blocks are available in the repository now (list attached)
15. Provision Engines via UI
- Capability for a user with appropriate privilege to provision an engine from a list of available templates set via Ops portal.
- Ability to pause & terminate the template. The engine provisioned successfully will show up in pipelines and SDK to run
16. Interim Payment checks & maximum engine burn rate
- Respective pre configured interim payment and maximum engine burn rate checks set for signups with credit card and without credit card
- Interim payment to trigger if your hourly burn rate is higher then the limit & all checks are passed during the time of provisioning engine and nightly cron job
17. Engine Transaction logs recording
- All engines provisioned for that tenant are recorded with transaction ID, the amount charged, state, who & when provisioned etc. and are listed under Billing
- Users with appropriate privilege can view those transaction logs and download it.
- Transaction logs across all tenants are listed in Ops Portal
18. Account Status
- Maintain various subscription account status in Chargify, Platform & Ops portal for tenants. Billing will vary basis the type of Status, the permissions, and privileges
- After getting into pipelines, you can see all your pipeline runs by clicking on the PIPELINE RUNS tab.
To get more details about a particular pipeline run like logs and metrics, you can click on that pipeline to see those details.