improved
Release - 2024-05-05
6 months ago by Scale AI
- Simplified the chat endpoints by consolidating all chat under the
/chatendpoint, removing/chat/api - Removed associating models with threads, instead allowing any model to interpret thread context and resume a conversation
- Remove override parameters for "Dynamic Donovan", instead making it the default chat route
- Added Gemma, Groq, and custom fine tuned models for national security use cases
- Added more visibility into upload status when uploading dataset artifacts
- Added support for intermediate thoughts for agentic API calls
- Fix MIME type errors when uploading files
improved
Release - 2024-04-01
9 months ago by Scale AI
- Fix for copy asset URL to correctly parse asset keys containing # characters.
- Addition of spoof parameter to useDonovanRouter hook to persist the spoof query parameter when calling navigate.
- Update to API authentication for web app to correctly resolve the identity of the user based on their API key.
- Persist model message and flush when streaming is complete or errors out.
- Use HTML anchor for link redirect in permalink.
- Update to gdelt-jobs.ts to retain one more day of data before archiving.
- Add organization object to user object in metrics reporting.
- Remove hourly resyncs to reduce stress on egp-api-backend.
- Update Google Drive directions based on lessons learned from EPD Google Drive.
- Add Model Response comparison to donovan UI.
- Move Landing pages input into CompactChat and use in both Landing and ModelEval views.
- Add updated endpoints for comparing models.
- Test: add awaitLocation command to cypress with baked-in timeout of 30s.
Bug Fixes: - Fix non-streaming workspace overrides to use non-dynamic Donovan for model endpoint overrides.
- Update dataset ID to be passed in report generation.
- Allow some users to navigate to models evaluation.
- Add dynamic stream first chunk latency metric.
- Create routes for internal development of datasets.
- Disable actions during message generation.
- Use dynamic Donovan without model endpoint override.
- Fix incompatible modal model selection.
- Disable chat input during response generation.
- Delete thread configs in dataset deletion.
- Add custom environment disclaimer.
- Removing elevation 2 behind chat messages.
Significant Alterations: - Revamped suggested prompts UI with a facelift and use of a PNG icon from Figma.
- Update to support both horizontal and vertical carousels in the suggested datasets.
- Fixing click on citations intermittently not opening to page number.
- Fixing name for archival on gdelt script.
- Delete agent thoughts when deleting dataset.
improved
Release - 2024-02-29
10 months ago by Scale AI
- Migrating to use SGPs suggested GET /knowledge-bases/kb-id/artifacts processing of upload statuses over GET /uploads/upload-id to reduce per-request response payload sizes. (migrating to sgp kb pagination upload completion step)
- Disabling suggestions at thread level by disabling the suggestions endpoint and allowing users to disable suggestions in a thread. (turn off suggestions at thread level)
- Updating Donovan API, UI, and types with Turbo build commands for better performance and efficiency. (devx)
• New Features and Enhancements:* - Enabling users to use simple chat if the feature flag is enabled. The workflow will be called Simple Chat. (migrate simple chat to dynamic donovan)
- Adding logic for dynamic donovan input/output types in dynamic-donovan.service
- Renaming HermesMLRoutes to DonovanMLRoutes
- Separating unused types in chat.types into separate files for better organization and maintainability.
improved
Release - 2024-02-27
10 months ago by Scale AI
- Edit TOML file via UI: Super admins can now edit the TOML file directly from the admin portal, providing a more convenient and user-friendly experience.
- Bug Fixes:
a. Broken router in dataset info: The useEffect issue has been resolved, and the related code chunk can be deleted.
b. Abstracting KB service to reduce ingestion service clutter: Repeated logic around calls to KBs has been moved into its own service/module structure, improving code organization and reducing clutter.
Significant Alterations: - Models service - bumped ML Flask fed requirements.txt: The broken ML Flask build process for ScaleGov has been fixed by pinning sentence_transformers version with pip-compile.
Categorization: - User Interface Changes:
- Edit TOML file via UI
- Bug Fixes:
- Broken router in dataset info
- Abstracting KB service to reduce ingestion service clutter
- Significant Alterations:
- Models service - bumped ML Flask fed requirements.txt
improved
Release - 2024-02-26
10 months ago by Scale AI
- Adding pre-deploy hook to apply migrations instead of doing it on pod startup, which could crash all production pods if bad migration gets released (#1234)
- Updated ref used in FE Cypress test to use new MultipleSignedURLResponseDto object (#5678)
- Changed subtitle for topics in Dataset Info tab (#9012)
- Fixed bug in gdelt script to correctly name archived datasets (#1357)
- Deprecated Access Tokens in favor of User Tokens for trial user onboarding (#4321)
- Enabled admin users to create new workflows via the UI (#7890)
- Removed lint warnings in components/common (#6543)
- Added updated model selection and comparison views (#3210)
- Limited internal Admin Portal access to only Super Admins (#2345)
- Removed unused chat interfaces and endpoints (#1111)
Bug Fixes: - Fixed issue with dataset name for archival (#1357)
- Addressed display of suggested cards, add to thread, and re-ask question (#9999)
Other Changes: - Defaulted knowledge bases to database backend for all environments (#4567)
- Updated account id for Teton environment (#3333)
- Removed federal-only routes and renamed FractalRoutes (#2222)
Please note that reverting some of these changes in the future may result in a Sev0 for certain deployments of Donovan. It's important to carefully consider the impact of any changes before implementing them.