Releases: LambdaLabsML/shaderunner
Releases · LambdaLabsML/shaderunner
Shaderunner v0.10.1
ShadeRunner Version 0.10.1
Security Updates:
This is only a security update, running pnpm audit --fix
.
Shaderunner v0.10.0
ShadeRunner Version 0.10.0
Features:
- changed
ada-002
->3-small
embedding method - summarization of hackernews comments
Shaderunner v0.9.0
ShadeRunner Version 0.9.0
Features:
- opening plugin does nothing -- let's user decide on mode to use
Refactor:
- streamlined IndexedDB database (using dexie)
Bugfixes:
- embedding not computed on startup
Shaderunner v0.8.1
ShadeRunner Version 0.8.0
Features:
- new feature: summarize all paragraphs on a page
- new feature: jump/scroll to first result after highlight has been applied
- defaults: find source / retrieval is now the default option (and much faster)
- automatic removal of old embedding databases
Minor:
- faster / more stable message passing of all plugin components
- simplified sidepanel ui appearance by hiding/showing switches depending on mode
- new active status: plugin is now active when sidepanel is open
- adapted background color to match new chrome version
- replaced localStorage with background script storage
Bugfixes:
- throttled calls to highlight (when user changes setting during computations)
- minified calls to highlight
- fixed error regarding missing class embeddings
Shaderunner v0.7.0
ShadeRunner Version 0.7.0
Features:
- Retrieval Mode readded
- Added retrieval-k slider for control of amount highlighted
- Control highlight quality with a "precision-vs-recall" slider
Minor:
- Added version to sidepanel and options tab
- styling:
- Added dark-mode style to options
- Adapted options page color
- Added logo to sidepanel
- Improved efficiency of mouse hover events in Legend
- Replaced useSessionStorage with useStorage for improved storage handling
- Updated global storage handling
- Testing:
- Implemented batchwise method in
llm_classify
- Implemented evaluation and visualization improvements in
tabs/testing
- Implemented batchwise method in
Bugfixes:
- Fixed visual bug for inactive topics in legend
- Fixed collapsible box in light-mode
- Bugfix for highlighting not starting without clicking a button
- Fixed scroller to exclude hidden elements
- Resolved issue with precise search limit not working
- Bugfix for amount highlighter not using score
- Addressed multiple highlight update issues in different modes
- Fixed issue with status display
- Corrected error handling in sidepanel when not connected to a webpage
- Bugfixes for issues with bad classifierData
Shaderunner v0.6.1
ShadeRunner Version 0.6.1
Minor:
- content extraction now searches till the end of the document
Bugfixes:
- content extraction bug: cannot trim undefined
- Testsethelper: did not save
document.title
- race conditions for status messages
- fixed embedding "commputing" status indicator (percentage was not showing up)
Shaderunner v0.6.0
ShadeRunner Version 0.6.0
Features:
- Reimplementation of core highlighting mechanism
- to increase quality uses precomputed fuzzy string distances on text nodes for matching
innerText
totextContent
- better text Node selection (only actually visible text nodes)
- to increase quality uses precomputed fuzzy string distances on text nodes for matching
- Performance:
- parallel embedding of classifier and content
- replaced ChromaDB by local IndexedDB
Minor:
- TestsetHelper: disabled any click events other than those of shaderunner in this mode
Bugfixes:
- Legend: "show all" after "focus on" did not show all highlights
- TestsetHelper: fixed interference with other highlighting modes
- LLM Classifier:
- fixed chat models
- using more stable class delimiter (
|
instead of,
)
- saved url was missing
window.location.search
Shaderunner v0.5.0
ShadeRunner Version 0.5.0
Features:
- Advanced/Amount Highlighted added: steer percentage of top highlights to highlight
- Advanced/ThoughtInfo added: shows thought process of llm (scope & reasoning)
- Advanced/Class Similarities added: force-spring plot to show similarities of classes on mouse-hover
- Suggest missing topic-buttton added to legend
Minor:
- Mouseover on highlights: now shows closest interesting and closest uninteresting class
- Options: obfuscate openai api key
- Legend: inactivate topics without any matches
- Sidebar: Adapted style of collapsible boxes + nested collapsible boxes
- Testing: started working on testing tab
- More Stable Highlighting: more sentences found by using
innerText
ondocument.body
Bugfixes:
- Highlighter:
- fixed caching of class embeddings (on sidebar reopen, on page refresh)
- fixed status indicator not updating for highlights
- fixed highlight reset not resetting all highlights
- fixed text finding when sentence starts later in a text node
- Background service worker:
- fixed "invalid URL" errors in various cases
- Legend:
- filterSelector not working
- fixed removal of topics not updating immediately
Shaderunner v0.4.0
ShadeRunner Version 0.4.0
Features:
- complete reimplementation / restructure of project (files, methods & functionality)
- complete redesign (with dark-mode!)
- plugin sidepanel instead of embedded controls
- batch-wise embedding
- improved speed of highlighting
- next/prev highlight buttons
Minor:
- added ui for outlier topics
- showing dynamic status of computations - show number of finds per class
- caching of class embeddings
- simple testset helper added
Code:
- many ts type mismatches fixed
- detached messaging from position of code using useGlobalStorage
- added boxicons icons
Shaderunner v0.3.0
ShadeRunner Version 0.3.0
Features:
- cache llm response & allow user to change topics
- improved string matching (first try by-word matching, then slower by-char matching)
- improved topic prompt by introducing concept of "interesting", "outliers" and "scope". also improved training examples
- new color for each positive topic added - added floating legend + click/hover effects to toggle highlight topics
- added focus mode (only preview, removes all but highlighted text)
Settings:
- ensure settings are set upon installation
- disable threshold values by default
- added gpt temperature
- added custom/openchat api endpoint
- added gpt-4 turbo preview
- disabled retrieval mode by default
Minor:
- mousehover on highlight shows assigned class and score
- update plugin icon status on tab change & startup
- reset highlights before next query
- use session storage for query data (also on a per-url basis)
- verbose-mode prints error when text not found on page
- cached embeddings not found sometimes
- dev-mode: enabled persistent storage for debugging
- collapsed view of thought process & histograms
- clear data on empty input
- slight visualization improvement to show when computations are ongoing
- input field more responsive
- increased highlight color variance
- changed default highlight class names (shaderunner-highlight
)
Bugfixes:
- prevent concurrent highlights on startup
- force plasmo-inline to be visible (fixes reddit) - fixed color not updating in some cases
- fixed retrieval text color
- storage quota exceed errors fixed (by moving scores from storage to state)
- do not highlight when not active
minimaleps
not working in all cases- enable partial classifierData objects in rendering
- highlight-update did not use async to call highlightUsingX methods