rixty
/
SillyTavern
mirror of https://github.com/SillyTavern/SillyTavern.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #2097 from Dakraid/feature/summarize-before-embedding 

Summarize the chat messages handed to the embedding beforehand
This commit is contained in:
Cohee 2024-04-18 22:49:34 +03:00 committed by GitHub
parent 0a54eeaf44 5e18b1872b
commit cc1c3f173c
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 154 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

128
public/scripts/extensions/vectors/index.js
View File

@ -1,5 +1,23 @@
import { eventSource, event_types, extension_prompt_types, getCurrentChatId, getRequestHeaders, is_send_press, saveSettingsDebounced, setExtensionPrompt, substituteParams } from '../../../script.js';
import { ModuleWorkerWrapper, extension_settings, getContext, modules, renderExtensionTemplateAsync } from '../../extensions.js';
import {
eventSource,
event_types,
extension_prompt_types,
getCurrentChatId,
getRequestHeaders,
is_send_press,
saveSettingsDebounced,
setExtensionPrompt,
substituteParams,
generateRaw,
} from '../../../script.js';
import {
ModuleWorkerWrapper,
extension_settings,
getContext,
modules,
renderExtensionTemplateAsync,
doExtrasFetch, getApiUrl,
} from '../../extensions.js';
import { collapseNewlines } from '../../power-user.js';
import { SECRET_KEYS, secret_state, writeSecret } from '../../secrets.js';
import { debounce, getStringHash as calculateHash, waitUntilCondition, onlyUnique, splitRecursive } from '../../utils.js';
@ -14,6 +32,10 @@ const settings = {
include_wi: false,
togetherai_model: 'togethercomputer/m2-bert-80M-32k-retrieval',
openai_model: 'text-embedding-ada-002',
summarize: false,
summarize_sent: false,
summary_source: 'main',
summary_prompt: '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 chats
enabled_chats: false,
@ -113,6 +135,56 @@ 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) {
for (const element of hashedMessages) {
element.text = await generateRaw(element.text, '', false, false, settings.summary_prompt);
}
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;
@ -135,14 +207,20 @@ 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 (newVectorItems.length > 0) {
const chunkedBatch = splitByChunks(newVectorItems.slice(0, batchSize));
console.log(`Vectors: Found ${newVectorItems.length} new items. Processing ${batchSize}...`);
await insertVectorItems(chatId, chunkedBatch);
}
@ -339,7 +417,7 @@ 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);
@ -348,7 +426,7 @@ 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) {
@ -389,13 +467,19 @@ 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()) {
if (message.mes) {
queryText += message.mes + '\n';
let hashedMessages = chat.map(x => ({ text: String(substituteParams(x.mes)) }));
if (settings.summarize && settings.summarize_sent) {
hashedMessages = await summarize(hashedMessages, settings.summary_source);
}
for (const message of hashedMessages.slice().reverse()) {
if (message.text) {
queryText += message.text + '\n';
i++;
}
@ -636,7 +720,7 @@ 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');
}
@ -757,6 +841,30 @@ jQuery(async () => {
saveSettingsDebounced();
});
$('#vectors_summarize').prop('checked', settings.summarize).on('input', () => {
settings.summarize = !!$('#vectors_summarize').prop('checked');
Object.assign(extension_settings.vectors, settings);
saveSettingsDebounced();
});
$('#vectors_summarize_user').prop('checked', settings.summarize_sent).on('input', () => {
settings.summarize_sent = !!$('#vectors_summarize_user').prop('checked');
Object.assign(extension_settings.vectors, settings);
saveSettingsDebounced();
});
$('#vectors_summary_source').val(settings.summary_source).on('change', () => {
settings.summary_source = String($('#vectors_summary_source').val());
Object.assign(extension_settings.vectors, settings);
saveSettingsDebounced();
});
$('#vectors_summary_prompt').val(settings.summary_prompt).on('input', () => {
settings.summary_prompt = String($('#vectors_summary_prompt').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);

41
public/scripts/extensions/vectors/settings.html
View File

@ -73,10 +73,12 @@
<input type="number" id="vectors_query" class="text_pole widthUnset" min="1" max="99" />
</div>
<label class="checkbox_label" for="vectors_include_wi" title="Query results can activate World Info entries.">
<input id="vectors_include_wi" type="checkbox" class="checkbox">
Include in World Info Scanning
</label>
<div class="flex-container">
<label class="checkbox_label expander" for="vectors_include_wi" title="Query results can activate World Info entries.">
<input id="vectors_include_wi" type="checkbox" class="checkbox">
Include in World Info Scanning
</label>
</div>
<hr>
@ -90,7 +92,6 @@
</label>
<div id="vectors_files_settings">
<div class="flex-container">
<div class="flex1" title="Only files past this size will be vectorized.">
<label for="vectors_size_threshold">
@ -123,6 +124,8 @@
Enabled for chat messages
</label>
<hr>
<div id="vectors_chats_settings">
<div id="vectors_advanced_settings">
<label for="vectors_template">
@ -165,6 +168,34 @@
<input type="number" id="vectors_insert" class="text_pole widthUnset" min="1" max="9999" />
</div>
</div>
<hr class="m-b-1">
<div class="flex-container flexFlowColumn">
<div class="flex-container alignitemscenter justifyCenter">
<i class="fa-solid fa-flask" title="Summarization for vectors is an experimental feature that may improve vectors or may worsen them. Use at your own discretion."></i>
<span>Vector Summarization</span>
</div>
<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 for vector generation
</label>
<i class="failure">Warning: This will slow down vector generation drastically, as all messages have to be summarized first.</i>
<label class="checkbox_label expander" for="vectors_summarize_user" title="Summarize sent chat messages before generating embeddings.">
<input id="vectors_summarize_user" type="checkbox" class="checkbox">
Summarize chat messages when sending
</label>
<i class="failure">Warning: This might cause your sent messages to take a bit to process and slow down response time.</i>
<label for="vectors_summary_source">Summarize with:</label>
<select id="vectors_summary_source" class="text_pole">
<option value="main">Main API</option>
<option value="extras">Extras API</option>
</select>
<label for="vectors_summary_prompt">Summary Prompt:</label>
<small>Only used when Main API is selected.</small>
<textarea id="vectors_summary_prompt" class="text_pole textarea_compact" rows="6" placeholder="This prompt will be sent to AI to request the summary generation."></textarea>
</div>
</div>
<small>
Old messages are vectorized gradually as you chat.