Dockery/Trilium
1
0
Fork 0
mirror of https://github.com/zadam/trilium.git synced 2025-11-03 20:06:08 +01:00
Code Issues Packages Projects Releases Activity

dynamically adjust context window sizes based on conversation context

Browse Source
This commit is contained in:
perf3ct
2025-03-30 22:13:40 +00:00
parent 29845c343c
commit f2cb013e14
10 changed files with 388 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
src/services/llm/pipeline/chat_pipeline.ts
View File

@@ -106,7 +106,8 @@ export class ChatPipeline {
// Get semantic context for regular queries
const semanticContext = await this.stages.semanticContextExtraction.execute({
noteId: input.noteId,
query: input.query
query: input.query,
messages: input.messages
});
context = semanticContext.context;
this.updateStageMetrics('semanticContextExtraction', contextStartTime);
@@ -136,10 +137,10 @@ export class ChatPipeline {
const llmStartTime = Date.now();
// Setup streaming handler if streaming is enabled and callback provided
const enableStreaming = this.config.enableStreaming &&
const enableStreaming = this.config.enableStreaming &&
modelSelection.options.stream !== false &&
typeof streamCallback === 'function';
if (enableStreaming) {
// Make sure stream is enabled in options
modelSelection.options.stream = true;
@@ -157,10 +158,10 @@ export class ChatPipeline {
await completion.response.stream(async (chunk: StreamChunk) => {
// Process the chunk text
const processedChunk = await this.processStreamChunk(chunk, input.options);
// Accumulate text for final response
accumulatedText += processedChunk.text;
// Forward to callback
await streamCallback!(processedChunk.text, processedChunk.done);
});
@@ -182,12 +183,12 @@ export class ChatPipeline {
const endTime = Date.now();
const executionTime = endTime - startTime;
// Update overall average execution time
this.metrics.averageExecutionTime =
this.metrics.averageExecutionTime =
(this.metrics.averageExecutionTime * (this.metrics.totalExecutions - 1) + executionTime) /
this.metrics.totalExecutions;
log.info(`Chat pipeline completed in ${executionTime}ms`);
return finalResponse;
@@ -235,12 +236,12 @@ export class ChatPipeline {
*/
private updateStageMetrics(stageName: string, startTime: number) {
if (!this.config.enableMetrics) return;
const executionTime = Date.now() - startTime;
const metrics = this.metrics.stageMetrics[stageName];
metrics.totalExecutions++;
metrics.averageExecutionTime =
metrics.averageExecutionTime =
(metrics.averageExecutionTime * (metrics.totalExecutions - 1) + executionTime) /
metrics.totalExecutions;
}
@@ -258,7 +259,7 @@ export class ChatPipeline {
resetMetrics(): void {
this.metrics.totalExecutions = 0;
this.metrics.averageExecutionTime = 0;
Object.keys(this.metrics.stageMetrics).forEach(stageName => {
this.metrics.stageMetrics[stageName] = {
totalExecutions: 0,