FIM strategy context improvements #2863
Conversation
…ecent operations.
… for FIM strategy
|
|
@beatlevic changed the base so it is easier to review |
| } | ||
|
|
||
| // Analyze and track global operations if we have enough history | ||
| if (item.history.length >= 2) { |
| * @param operation The operation to add | ||
| * @param filepath The file where the operation occurred | ||
| */ | ||
| private addGlobalOperation(operation: UserAction, filepath: string): void { |
There was a problem hiding this comment.
this needs to respect kilocodeignore/gitignore, see #2852
| * Formula: |A ∩ B| / |A ∪ B| | ||
| * Where A and B are sets of symbols from each string | ||
| */ | ||
| export function jaccardSimilarity(a: string, b: string): number { |
There was a problem hiding this comment.
this feels like it could be slow for large files
| let intersection = 0 | ||
| for (const symbol of aSet) { | ||
| if (bSet.has(symbol)) { | ||
| intersection++ |
There was a problem hiding this comment.
we can't use these?
https://viljams.medium.com/new-javascript-set-methods-are-here-b21e9a37bc4b
| export function jaccardSimilarity(a: string, b: string): number { | ||
| const aSet = getSymbolsForSnippet(a) | ||
| const bSet = getSymbolsForSnippet(b) | ||
| const union = new Set([...aSet, ...bSet]).size |
There was a problem hiding this comment.
if we know the size of the intersection, we know the size of the union right? So this isn't necessary (and probably slow, especially since you're converting to an array in between)
|
|
||
| // Get window around cursor for similarity comparison | ||
| const position = context.range.start | ||
| const windowSize = 500 // characters before and after cursor |
There was a problem hiding this comment.
in the context of an extension/site windowSize seems like a visual thing, maybe characterLookAroundSize or something?
| /** | ||
| * Deduplicate snippets from the same file by merging overlapping content | ||
| */ | ||
| export function deduplicateSnippets(snippets: RankedSnippet[]): RankedSnippet[] { |
There was a problem hiding this comment.
why are these and following methods unused? Especially I think the constraining of amount of syntax is especially important as it is easy to overwhelm small models. We might even need to compress / slice the current file a bit if it is too large
| */ | ||
| export function fillPromptWithSnippets( | ||
| snippets: RankedSnippet[], | ||
| maxTokens: number, |
There was a problem hiding this comment.
we probably need to use the current file as input for maxTokens
2 participants
Adds the following to the FIM strategy context: