The Backyard Summer Special is here!
Try the Unlimited Free Plan!
Changelog
Latest Updates & Improvements to Backyard AI
0.30.0
Oct 16, 2024
stable
- Bug fixes
- Fixed issue where uppercase {User} and {Character} in character settings would not be replaced with the actual user/character name
- Added undo button under the last message
- Fixed long-press menu not showing on the Android app
0.29.2
Oct 15, 2024
stable
- Chat Page Updates
- You can now add a full-height background image to any chat and upload them to the Hub alongside your characters
- Added chat bubbles and full width chat display modes (toggleable in the settings)
- Moved author's note and impersonate buttons below the input box
- Moved model selection to a dedicated tab on the chat settings page
- Added right click menu for undo and upvote actions
- Added Cmd+Enter shortcut to save author’s note and messages edits
- Chat-specific settings
- First shipped in version 0.28.4, now available for all non-beta clients
- Chats are now independent of each other and can be saved with different settings
- StyleTTS2 Voice Model
- Added 24 new Character voices powered by StyleTTS2, an advanced model for human-level text-to-speech
- Updated UI for voice selection
- Voice selection can now be uploaded to the Hub alongside your characters
- To use StyleTTS2, you will need to sign in to your account on web, mobile, or desktop (not available locally/offline)
- Piper voices will still be available for offline use
- Character Hub Upload Improvements
- Ability to select which chat settings to include with Character upload
- Improved visual highlighting for selecting tags
- New “Experimental” Backend
- First shipped in version 0.28.22, now available for all non-beta clients
- Improvements to token rate and prompt processing on some machines
- Vulkan optimizations
- Nemotron/Minitron model support
- Gemma2 flash attention support except on Vulkan
- Other Improvements
- You can edit banned hub characters and resubmit them for approval
- 你好; こんにちは; Bonjour; Google Translate is now supported on the web for chat messages
- Increases maximum response length to 384 tokens
- Added button to delete account on the Settings page
- Improved Windows header dragging
- Added ability to sort user personas in the App Settings
- Mobile App Updates
- Added ability to swipe to regenerate a response
- Moved edit button and other menus to long-press popup menu
- Performance improvements across the app
- Editing messages is should be much easier now
- Scrolling issues on message are fixed now
- UI Toasts have been updates to be larger so the message is not cut off
- Light mode is now supported
- Dozens of other stability improvements
- Bug fixes
- Fixed Desktop shortcut being created every update on Windows
- Fixed model download progress not updating while chatting
- Fixed issue where some characters would not appear on the home page
- Fixed issue where uploading images for Characters from the desktop app would not appear in the Hub
- Fixed bug causing broken message streams
- Fixed "invalid state transition" errors on desktop chat page
- Fixed issue where the chat dropdown list was not scrollable
- Fixed issue where tethering toggle would not save on mobile
- Fixed issue where character images would not display on tethering
- Fixed accounts that could not create characters
- Fixed auto-speech starting after "Write for Me" finishes
- Fixed {char} and {user} being swapped on character import/export
- Fixed out of order chat imports
- Fixed blinking cursor during "Write for Me" response
- Fixed "invalid state transition" errors
- Fixed issue preventing certain chats from loading
- Fixed issue where creating a new chat would reset model & character settings
- Fixed issue where the character changes were not reflected correctly in the UI after saving
- Fixed issue where the user persona could not be changed on a character
- Fixed issue where the "Docs" button in the error page was not working
- Fixed "Primary chat not found" error message
- Fixed issue on ios when opening characters on the Hub
- Fixed out of order "other chats" in the chat page dropdown
Updates from 23 beta releases included
0.28.0
Aug 29, 2024
stable
- Improvements
- Updated the internal data model to support upcoming features and improvements
- User personas created on the General Settings page are now shared across Characters. If you update a persona, it will update for all Characters using that persona.
- Improved the Character Hub UI
- Minor color palette update across dark/light mode
- Pro/Advanced/Standard plan users are prioritized over Free plan users in the message queue
- When a cloud model is deprecated, there is now a dropdown to select a replacement model
- Improved error messages for malformed models
- Bug fixes
- Fixed issue where dark/light mode preference would not save
- Fixed sidebar overflow on small screens
- Fixed issue where changing the folder name in the sidebar would not reflect on the homepage
- Fixed missing "Backyard Bot" default image on the chat page
- When loading a Character, the most recently used chat is now selected
- Fixed Character settings UI not updating after edits in the mobile app
- Fixed prompting issues for Mistral-Instruct
- Fixed MLock settings not saving
- Fixed slowdowns on Nvidia cards ("Experimental" backend only)
- Fixed model initialization slowdowns on Vulkan backends
Updates from 7 beta releases included
0.27.0
Aug 26, 2024
stable
- Bug fixes
- Fixed prompt regression from the last update
- Fixed Mistral-Instruct prompt templating
- Improvements
- Added support for Command-R and Mistral-Instruct prompt templates
Updates from 6 beta releases included
0.26.0
Aug 5, 2024
stable
- Backyard Cloud - Summer Special Promotion
- Free plan: Unlimited messages for Fimbulvetr 10.7B model
- Referral program (see blog post)
- iOS & Android apps are now in public beta (see Discord)
- Bug Fixes & Improvements
- Fixed user persona images not updating
- Fixed incorrect limit for Author’s Note text input
- Fixed issue where subfolders would be hidden when deleting a parent folder
- Fixed app updates hanging at 99%
- Fixed various issues relating to homepage folders
- Fixed model initialization failing on Intel Macs
- New “Experimental” Backend
- Mistral-NeMo support
- Performance improvements across all GPU types
- Hotfix
- Fixed chats not loading for certain models (DarkForest)
- Character folder bug fixes
- Fixed chat startup issues on Intel Mac machines
- Fixed "Your system is missing the requisite DLL(s)." error
- Disabled experimental bins on MacOS that were preventing app startup
Updates from 27 beta releases included
0.25.0
Jul 24, 2024
stable
- Homepage Folders & Custom Sorting
- Create folders on the homepage
- Sort characters via drag & drop
- Mobile App & Mobile Web Updates
- Improved mobile chat page UI
- UI updates to better match desktop app
- General performance improvements for mobile
- Better hub navigation
- Search improvements on mobile
- Bug Fixes & Improvements
- Fixed issue with cursor jumping to the end of text inputs
- Added Gemma2 prompt template
- Fixed issue causing Windows app to crash while tethering
- Fixed Gemma2 on Windows
- Support wildcards in lore keys
- Updated "Experimental" backend to support Gemma models
- Fixed inaccurate example dialogue token count
- Fixed broken button to disconnect tethering
- Improved Windows desktop app UI
Updates from 5 beta releases included
0.24.0
Jul 3, 2024
stable
- Added Status Page on Web
- We now have a public status page that shows the current uptime of the web app, character hub, and our cloud models
- It also includes the average wait time and tokens/second for all cloud models
- Bug Fixes & Improvements
- Speedup of message response time when using context sizes above 6k
- Various fixes for mobile tethering
- Fixed issue where uploading/downloading characters images to/from the hub would be out of order
- Fixed issue where GIFs were not correctly displayed
- Fixed issue with chat message edits not saving
- Fixed User personas not saving correctly
- Fixed "Minor Version Updates" toggle on the settings page
- Fixed issue where chosen Avatar image would not stay selected
- Fixed incorrect Avatar image dimensions on download
- Fixed issue where updating Characters would break images
- Character avatar image carousel order is remembered when navigating back to the Chat Page
- Fixed incorrect image dimensions after downloading a character
- Fixed issue where images wouldn't correctly download from the Character Hub
- Fixed issue where the "Minor Version Updates" settings toggle state wasn't correctly reflected
- Fixed rare render issues with dropdown menu
- Fixed issue where deleting a character would delete a saved persona's image
- Fixed issue where editing responses occasionally deleted the response on save
Updates from 5 beta releases included
0.23.2
Jun 24, 2024
stable
- Character Page on the Hub
- UI updates to the Character Pages on the Hub
- Bug Fixes & Improvements
- Fixed bug causing the app homepage to load slowly when the app has over 100 characters
- Fixed bug causing stop button to break when using the "Experimental" backend
- Improved loading times for large models
- Fixed 'cublas_12.1.0_avx2-v0.2... has stopped working' error on Windows
- Fixed issue causing image controls to jump around on characters with multiple different-sized images
0.23.0 - 0.23.1 Internal
0.22.3
Jun 21, 2024
stable
- Bug Fixes & Improvements
- Fixed under allocation of VRAM on some Windows devices
- Fixed "Client Stale Error" on the Desktop chat page
- Fixed issue with CUDA device detection
- Improved model loading speeds on CUDA
- Removed Basic mode from character creation to support upcoming features
- Fixed issue with uploading characters to the Hub
- Fixed IP rate limiting issue on Free Plan
- Text-to-Speech fixed on web
- Fixed occasional 400 errors on long chats using Cloud models
0.22.0 - 0.22.2 Internal
0.21.0
Jun 17, 2024
stable
- Backyard Cloud Revamp
- Free plan - increased quota on Fimbulvetr to 50 free messages per day (up from 100 per week)
- Every registered user receives 25 message credits per week that can be applied to any Cloud model (including Pro)
- Promoted Fimbulvetr to the default model on Free and Standard plans
- Added new "Advanced" subscription plan
- Added Fimbulvetr 10.7B at 8,192 context tokens to Advanced and Pro plans
- Standard plan - Mythomax-Kimiko 13B added at 4,096 context tokens
- Pro plan - Llama 3 Jamet MK.V Blackroot 8B added at 10,240 context tokens
- Pro plan - Midnight-Rose 70B upgraded to a higher quality format
- Pro plan - Psyonic-Cetacean 20B replaced with Psyonic-Cetacean Ultra 20B
- Pro plan - early access to iOS app is now available
- Advanced plan - Chunky Lemon-Cookie 11B added at 8,192 context tokens
- Desktop App Updates
- Added ability to opt out of minor release auto-updates in the settings
- Updated Stable backend to support IQ quants
- Fixed bad outputs on certain Mac devices using Apple Metal
- Respect "iogpu.wired_limit_mb" setting on Apple devices
- Better auto-GPU detection for large parameter models
- Modest performance improvements for CUDA and AMD
- Added support for quantized KV cache on Experimental Backend (Nvidia GPUs only)
- CLBlast will begin to be phased out of future releases (use Vulkan instead)
- App Improvements
- Added input field caching on Character creation page to prevent unsaved edits from being lost
- Added popup dialogue to disable undo confirmation
- Significantly improved load times across the Character Hub
- Fixed trending algorithm breaking on the Character Hub
- Added more descriptive error messages to chat page
- Added login options for Apple and Twitter/X
- Added ability to upload characters from the web app
- Lore item keys can now be up to 300 characters in length
- During lore injection, only the matching key is inserted into context
Tethering: Disabled
No model loaded