mincomk/risuai
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: improve HypaV3 (#721)

Browse Source 
# PR Checklist
- [x] Have you checked if it works normally in all models? *Ignore this
if it doesn't use models.*
- [x] Have you checked if it works normally in all web, local, and node
hosted versions? If it doesn't, have you blocked it in those versions?
- [x] Have you added type definitions?

# Description
- feat: add translate button to HypaV3 modal
- feat: add dual-action translation (cached/regenerate) in HypaV3 modal
- Implements shift+click (desktop) and double tap (mobile) for
regenerating translations while maintaining regular click/tap for cached
translations.
- feat: improve ratio sliders interaction in HypaV3 settings
- feat: add lastSelectedSummaries property to store selected memory
indices
- feat: display next message to be summarized in HypaV3 modal
This commit is contained in:
kwaroran
2025-01-17 22:54:08 +09:00
committed by GitHub
parent 847a1553bb b43eec49d1
commit e073d66bca
5 changed files with 707 additions and 288 deletions
Download Patch File Download Diff File Expand all files Collapse all files

229
src/ts/process/memory/hypav3.ts
View File

@@ -24,6 +24,7 @@ interface Summary {
interface HypaV3Data {
summaries: Summary[];
lastSelectedSummaries?: number[];
}
export interface SerializableHypaV3Data {
@@ -32,6 +33,7 @@ export interface SerializableHypaV3Data {
chatMemos: string[];
isImportant: boolean;
}[];
lastSelectedSummaries?: number[];
}
interface SummaryChunk {
@@ -39,6 +41,10 @@ interface SummaryChunk {
summary: Summary;
}
const minChatsForSimilarity = 3;
const maxSummarizationFailures = 3;
const summarySeparator = "\n\n";
// Helper function to check if one Set is a subset of another
function isSubset(subset: Set<string>, superset: Set<string>): boolean {
for (const elem of subset) {
@@ -51,6 +57,7 @@ function isSubset(subset: Set<string>, superset: Set<string>): boolean {
function toSerializableHypaV3Data(data: HypaV3Data): SerializableHypaV3Data {
return {
...data,
summaries: data.summaries.map((summary) => ({
...summary,
chatMemos: [...summary.chatMemos],
@@ -60,6 +67,7 @@ function toSerializableHypaV3Data(data: HypaV3Data): SerializableHypaV3Data {
function toHypaV3Data(serialData: SerializableHypaV3Data): HypaV3Data {
return {
...serialData,
summaries: serialData.summaries.map((summary) => ({
...summary,
// Convert null back to undefined (JSON serialization converts undefined to null)
@@ -87,14 +95,17 @@ function cleanOrphanedSummary(chats: OpenAIChat[], data: HypaV3Data): void {
const removedCount = originalLength - data.summaries.length;
if (removedCount > 0) {
console.log(`[HypaV3] Cleaned ${removedCount} orphaned summaries`);
console.log(`[HypaV3] Cleaned ${removedCount} orphaned summaries.`);
}
}
export async function summarize(
stringifiedChats: string
oaiChats: OpenAIChat[]
): Promise<{ success: boolean; data: string }> {
const db = getDatabase();
const stringifiedChats = oaiChats
.map((chat) => `${chat.role}: ${chat.content}`)
.join("\n");
if (db.supaModelType === "distilbart") {
try {
@@ -103,7 +114,7 @@ export async function summarize(
} catch (error) {
return {
success: false,
data: "[HypaV3] " + error,
data: error,
};
}
}
@@ -116,7 +127,7 @@ export async function summarize(
switch (db.supaModelType) {
case "instruct35": {
console.log(
"[HypaV3] Using openAI gpt-3.5-turbo-instruct for summarization"
"[HypaV3] Using openAI gpt-3.5-turbo-instruct for summarization."
);
const requestPrompt = `${stringifiedChats}\n\n${summarizePrompt}\n\nOutput:`;
@@ -165,7 +176,7 @@ export async function summarize(
}
case "subModel": {
console.log(`[HypaV3] Using ax model ${db.subModel} for summarization`);
console.log(`[HypaV3] Using ax model ${db.subModel} for summarization.`);
const requestMessages: OpenAIChat[] = parseChatML(
summarizePrompt.replaceAll("{{slot}}", stringifiedChats)
@@ -190,14 +201,17 @@ export async function summarize(
"memory"
);
if (
response.type === "fail" ||
response.type === "streaming" ||
response.type === "multiline"
) {
if (response.type === "streaming" || response.type === "multiline") {
return {
success: false,
data: "Unexpected response type",
data: "unexpected response type",
};
}
if (response.type === "fail") {
return {
success: false,
data: response.result,
};
}
@@ -207,12 +221,43 @@ export async function summarize(
default: {
return {
success: false,
data: `Unsupported model ${db.supaModelType} for summarization`,
data: `unsupported model ${db.supaModelType} for summarization`,
};
}
}
}
async function retryableSummarize(
oaiChats: OpenAIChat[]
): Promise<{ success: boolean; data: string }> {
let summarizationFailures = 0;
while (summarizationFailures < maxSummarizationFailures) {
console.log(
"[HypaV3] Attempting summarization:",
"\nAttempt:",
summarizationFailures + 1,
"\nTarget:",
oaiChats
);
const summarizeResult = await summarize(oaiChats);
if (!summarizeResult.success) {
console.log("[HypaV3] Summarization failed:", summarizeResult.data);
summarizationFailures++;
if (summarizationFailures >= maxSummarizationFailures) {
return summarizeResult;
}
continue;
}
return summarizeResult;
}
}
export async function hypaMemoryV3(
chats: OpenAIChat[],
currentTokens: number,
@@ -226,9 +271,6 @@ export async function hypaMemoryV3(
error?: string;
memory?: SerializableHypaV3Data;
}> {
const minChatsForSimilarity = 3;
const maxSummarizationFailures = 3;
const summarySeparator = "\n\n";
const db = getDatabase();
// Validate settings
@@ -250,6 +292,7 @@ export async function hypaMemoryV3(
// Load existing hypa data if available
let data: HypaV3Data = {
summaries: [],
lastSelectedSummaries: [],
};
if (room.hypaV3Data) {
@@ -360,8 +403,8 @@ export async function hypaMemoryV3(
i,
"\nRole:",
chat.role,
"\nContent:\n",
chat.content,
"\nContent:",
"\n" + chat.content,
"\nTokens:",
chatTokens
);
@@ -387,53 +430,29 @@ export async function hypaMemoryV3(
currentTokens - toSummarizeTokens < targetTokens
) {
console.log(
`[HypaV3] Stopping summarization: would reduce below target tokens (${currentTokens} - ${toSummarizeTokens} < ${targetTokens})`
`[HypaV3] Stopping summarization: currentTokens(${currentTokens}) - toSummarizeTokens(${toSummarizeTokens}) < targetTokens(${targetTokens})`
);
break;
}
// Attempt summarization
let summarizationFailures = 0;
const stringifiedChats = toSummarize
.map((chat) => `${chat.role}: ${chat.content}`)
.join("\n");
const summarizeResult = await retryableSummarize(toSummarize);
while (summarizationFailures < maxSummarizationFailures) {
console.log(
"[HypaV3] Attempting summarization:",
"\nAttempt:",
summarizationFailures + 1,
"\nTarget:",
toSummarize
);
const summarizeResult = await summarize(stringifiedChats);
if (!summarizeResult.success) {
console.log("[HypaV3] Summarization failed:", summarizeResult.data);
summarizationFailures++;
if (summarizationFailures >= maxSummarizationFailures) {
return {
currentTokens,
chats,
error: "[HypaV3] Summarization failed after maximum retries",
memory: toSerializableHypaV3Data(data),
};
}
continue;
}
data.summaries.push({
text: summarizeResult.data,
chatMemos: new Set(toSummarize.map((chat) => chat.memo)),
isImportant: false,
});
break;
if (!summarizeResult.success) {
return {
currentTokens,
chats,
error: `[HypaV3] Summarization failed after maximum retries: ${summarizeResult.data}`,
memory: toSerializableHypaV3Data(data),
};
}
data.summaries.push({
text: summarizeResult.data,
chatMemos: new Set(toSummarize.map((chat) => chat.memo)),
isImportant: false,
});
currentTokens -= toSummarizeTokens;
startIdx = endIdx;
}
@@ -529,9 +548,9 @@ export async function hypaMemoryV3(
selectedRecentSummaries.length,
"\nSummaries:",
selectedRecentSummaries,
"\nReserved Recent Memory Tokens:",
"\nReserved Tokens:",
reservedRecentMemoryTokens,
"\nConsumed Recent Memory Tokens:",
"\nConsumed Tokens:",
consumedRecentMemoryTokens
);
}
@@ -608,56 +627,33 @@ export async function hypaMemoryV3(
// (2) Summarized recent chat search
if (db.hypaV3Settings.enableSimilarityCorrection) {
let summarizationFailures = 0;
// Attempt summarization
const recentChats = chats.slice(-minChatsForSimilarity);
const stringifiedRecentChats = recentChats
.map((chat) => `${chat.role}: ${chat.content}`)
.join("\n");
const summarizeResult = await retryableSummarize(recentChats);
while (summarizationFailures < maxSummarizationFailures) {
console.log(
"[HypaV3] Attempting summarization:",
"\nAttempt:",
summarizationFailures + 1,
"\nTarget:",
recentChats
);
const summarizeResult = await summarize(stringifiedRecentChats);
if (!summarizeResult.success) {
console.log("[HypaV3] Summarization failed:", summarizeResult.data);
summarizationFailures++;
if (summarizationFailures >= maxSummarizationFailures) {
return {
currentTokens,
chats,
error: "[HypaV3] Summarization failed after maximum retries",
memory: toSerializableHypaV3Data(data),
};
}
continue;
}
const searched = await processor.similaritySearchScoredEx(
summarizeResult.data
);
for (const [chunk, similarity] of searched) {
const summary = chunk.summary;
scoredSummaries.set(
summary,
(scoredSummaries.get(summary) || 0) + similarity
);
}
console.log("[HypaV3] Similarity corrected");
break;
if (!summarizeResult.success) {
return {
currentTokens,
chats,
error: `[HypaV3] Summarization failed after maximum retries: ${summarizeResult.data}`,
memory: toSerializableHypaV3Data(data),
};
}
const searched = await processor.similaritySearchScoredEx(
summarizeResult.data
);
for (const [chunk, similarity] of searched) {
const summary = chunk.summary;
scoredSummaries.set(
summary,
(scoredSummaries.get(summary) || 0) + similarity
);
}
console.log("[HypaV3] Similarity corrected.");
}
// Sort in descending order
@@ -677,6 +673,8 @@ export async function hypaMemoryV3(
"[HypaV3] Trying to add similar summary:",
"\nSummary Tokens:",
summaryTokens,
"\nConsumed Similar Memory Tokens:",
consumedSimilarMemoryTokens,
"\nReserved Tokens:",
reservedSimilarMemoryTokens,
"\nWould exceed:",
@@ -689,7 +687,7 @@ export async function hypaMemoryV3(
reservedSimilarMemoryTokens
) {
console.log(
`[HypaV3] Stopping similar memory selection: would exceed reserved tokens (${consumedSimilarMemoryTokens} + ${summaryTokens} > ${reservedSimilarMemoryTokens})`
`[HypaV3] Stopping similar memory selection: consumedSimilarMemoryTokens(${consumedSimilarMemoryTokens}) + summaryTokens(${summaryTokens}) > reservedSimilarMemoryTokens(${reservedSimilarMemoryTokens})`
);
break;
}
@@ -706,9 +704,9 @@ export async function hypaMemoryV3(
selectedSimilarSummaries.length,
"\nSummaries:",
selectedSimilarSummaries,
"\nReserved Similar Memory Tokens:",
"\nReserved Tokens:",
reservedSimilarMemoryTokens,
"\nConsumed Similar Memory Tokens:",
"\nConsumed Tokens:",
consumedSimilarMemoryTokens
);
}
@@ -770,9 +768,9 @@ export async function hypaMemoryV3(
selectedRandomSummaries.length,
"\nSummaries:",
selectedRandomSummaries,
"\nReserved Random Memory Tokens:",
"\nReserved Tokens:",
reservedRandomMemoryTokens,
"\nConsumed Random Memory Tokens:",
"\nConsumed Tokens:",
consumedRandomMemoryTokens
);
}
@@ -814,10 +812,15 @@ export async function hypaMemoryV3(
if (currentTokens > maxContextTokens) {
throw new Error(
`[HypaV3] Unexpected input token count:\nCurrent Tokens:${currentTokens}\nMax Context Tokens:${maxContextTokens}`
`[HypaV3] Unexpected error: input token count (${currentTokens}) exceeds max context size (${maxContextTokens})`
);
}
// Save last selected summaries
data.lastSelectedSummaries = selectedSummaries.map((selectedSummary) =>
data.summaries.findIndex((summary) => summary === selectedSummary)
);
const newChats: OpenAIChat[] = [
{
role: "system",
@@ -854,7 +857,7 @@ class HypaProcesserEx extends HypaProcesser {
summaryChunkVectors: SummaryChunkVector[] = [];
// Calculate dot product similarity between two vectors
similarity(a: VectorArray, b: VectorArray) {
similarity(a: VectorArray, b: VectorArray): number {
let dot = 0;
for (let i = 0; i < a.length; i++) {
@@ -864,7 +867,7 @@ class HypaProcesserEx extends HypaProcesser {
return dot;
}
async addSummaryChunks(chunks: SummaryChunk[]) {
async addSummaryChunks(chunks: SummaryChunk[]): Promise<void> {
// Maintain the superclass's caching structure by adding texts
const texts = chunks.map((chunk) => chunk.text);