Add extras summarization API, fix missing substitution of placeholders

public/scripts/extensions/vectors/index.js

@@ -26,6 +26,7 @@ const settings = {
     togetherai_model: 'togethercomputer/m2-bert-80M-32k-retrieval',
     openai_model: 'text-embedding-ada-002',
     summarize: false,
+    summary_source: 'main',
 
     // For chats
     enabled_chats: false,
@@ -125,6 +126,57 @@ function splitByChunks(items) {
     return chunkedItems;
 }
 
+async function summarizeExtra(hashedMessages) {
+    for (const element of hashedMessages) {
+        try {
+            const url = new URL(getApiUrl());
+            url.pathname = '/api/summarize';
+
+            const apiResult = await doExtrasFetch(url, {
+                method: 'POST',
+                headers: {
+                    'Content-Type': 'application/json',
+                    'Bypass-Tunnel-Reminder': 'bypass',
+                },
+                body: JSON.stringify({
+                    text: element.text,
+                    params: {},
+                }),
+            });
+
+            if (apiResult.ok) {
+                const data = await apiResult.json();
+                element.text = data.summary;
+            }
+        }
+        catch (error) {
+            console.log(error);
+        }
+    }
+
+    return hashedMessages;
+}
+
+async function summarizeMain(hashedMessages) {
+    const sysPrompt = 'Pause your roleplay. Summarize the most important parts of the message. Limit yourself to 250 words or less. Your response should include nothing but the summary.';
+    for (const element of hashedMessages) {
+        element.text = await generateRaw(element.text, '', false, false, sysPrompt);
+    }
+
+    return hashedMessages;
+}
+
+async function summarize(hashedMessages, endpoint = 'main') {
+    switch (endpoint) {
+        case 'main':
+            return await summarizeMain(hashedMessages);
+        case 'extras':
+            return await summarizeExtra(hashedMessages);
+        default:
+            console.error('Unsupported endpoint', endpoint);
+    }
+}
+
 async function synchronizeChat(batchSize = 5) {
     if (!settings.enabled_chats) {
         return -1;
@@ -147,21 +199,20 @@ async function synchronizeChat(batchSize = 5) {
             return -1;
         }
 
-        const hashedMessages = context.chat.filter(x => !x.is_system).map(x => ({ text: String(x.mes), hash: getStringHash(x.mes), index: context.chat.indexOf(x) }));
+        let hashedMessages = context.chat.filter(x => !x.is_system).map(x => ({ text: String(substituteParams(x.mes)), hash: getStringHash(substituteParams(x.mes)), index: context.chat.indexOf(x) }));
         const hashesInCollection = await getSavedHashes(chatId);
 
+        if (settings.summarize) {
+            hashedMessages = await summarize(hashedMessages, settings.summary_source);
+        }
+
         const newVectorItems = hashedMessages.filter(x => !hashesInCollection.includes(x.hash));
         const deletedHashes = hashesInCollection.filter(x => !hashedMessages.some(y => y.hash === x));
 
-        if (settings.summarize) {
-            const sysPrompt = 'Pause your roleplay. Summarize the most important parts of the message. Limit yourself to 250 words or less. Your response should include nothing but the summary.';
-            for (const element of hashedMessages) {
-                element.text = await generateRaw(element.text, '', false, false, sysPrompt);
-            }
-        }
 
         if (newVectorItems.length > 0) {
             const chunkedBatch = splitByChunks(newVectorItems.slice(0, batchSize));
+
             console.log(`Vectors: Found ${newVectorItems.length} new items. Processing ${batchSize}...`);
             await insertVectorItems(chatId, chunkedBatch);
         }
@@ -358,7 +409,7 @@ async function rearrangeChat(chat) {
             if (retainMessages.includes(message) || !message.mes) {
                 continue;
             }
-            const hash = getStringHash(message.mes);
+            const hash = getStringHash(substituteParams(message.mes));
             if (queryHashes.includes(hash) && !insertedHashes.has(hash)) {
                 queriedMessages.push(message);
                 insertedHashes.add(hash);
@@ -367,7 +418,7 @@ async function rearrangeChat(chat) {
 
         // Rearrange queried messages to match query order
         // Order is reversed because more relevant are at the lower indices
-        queriedMessages.sort((a, b) => queryHashes.indexOf(getStringHash(b.mes)) - queryHashes.indexOf(getStringHash(a.mes)));
+        queriedMessages.sort((a, b) => queryHashes.indexOf(getStringHash(substituteParams(b.mes))) - queryHashes.indexOf(getStringHash(substituteParams(a.mes))));
 
         // Remove queried messages from the original chat array
         for (const message of chat) {
@@ -408,13 +459,19 @@ const onChatEvent = debounce(async () => await moduleWorker.update(), 500);
  * @param {object[]} chat Chat messages
  * @returns {string} Text to query
  */
-function getQueryText(chat) {
+async function getQueryText(chat) {
     let queryText = '';
     let i = 0;
 
-    for (const message of chat.slice().reverse()) {
+    let hashedMessages = chat.map(x => ({ text: String(substituteParams(x.mes)) }));
-        if (message.mes) {
+
-            queryText += message.mes + '\n';
+    if (settings.summarize) {
+        hashedMessages = await summarize(hashedMessages, settings.summary_source);
+    }
+
+    for (const message of hashedMessages.slice().reverse()) {
+        if (message.text) {
+            queryText += message.text + '\n';
             i++;
         }
 
@@ -655,7 +712,7 @@ async function onViewStatsClick() {
 
     const chat = getContext().chat;
     for (const message of chat) {
-        if (hashesInCollection.includes(getStringHash(message.mes))) {
+        if (hashesInCollection.includes(getStringHash(substituteParams(message.mes)))) {
             const messageElement = $(`.mes[mesid="${chat.indexOf(message)}"]`);
             messageElement.addClass('vectorized');
         }
@@ -782,6 +839,12 @@ jQuery(async () => {
         saveSettingsDebounced();
     });
 
+    $('#vectors_summary_source').val(settings.togetherai_model).on('change', () => {
+        settings.summary_source = String($('#vectors_summary_source').val());
+        Object.assign(extension_settings.vectors, settings);
+        saveSettingsDebounced();
+    });
+
     $('#vectors_message_chunk_size').val(settings.message_chunk_size).on('input', () => {
         settings.message_chunk_size = Number($('#vectors_message_chunk_size').val());
         Object.assign(extension_settings.vectors, settings);

public/scripts/extensions/vectors/settings.html

@@ -78,10 +78,20 @@
                     <input id="vectors_include_wi" type="checkbox" class="checkbox">
                     Include in World Info Scanning
                 </label>
+            </div>
+
+            <hr>
+
+            <div class="flex-container flexFlowColumn">
                 <label class="checkbox_label expander" for="vectors_summarize" title="Summarize chat messages before generating embeddings.">
                     <input id="vectors_summarize" type="checkbox" class="checkbox">
                     Summarize chat messages
                 </label>
+                <label for="vectors_summary_source">Summarize with:</label>
+                <select id="vectors_summary_source">
+                    <option value="main">Main API</option>
+                    <option value="extras">Extras API</option>
+                </select>
             </div>
 
             <hr>