mirror of
https://github.com/SillyTavern/SillyTavern.git
synced 2025-06-05 21:59:27 +02:00
Merge branch 'staging' into separate-kobold-endpoints
This commit is contained in:
@ -2629,12 +2629,12 @@ class StreamingProcessor {
|
||||
|
||||
if (!isImpersonate && !isContinue && Array.isArray(this.swipes) && this.swipes.length > 0) {
|
||||
for (let i = 0; i < this.swipes.length; i++) {
|
||||
this.swipes[i] = cleanUpMessage(this.removePrefix(this.swipes[i]), false, false, true, !isFinal);
|
||||
this.swipes[i] = cleanUpMessage(this.removePrefix(this.swipes[i]), false, false, true, this.stoppingStrings);
|
||||
}
|
||||
}
|
||||
|
||||
text = this.removePrefix(text);
|
||||
let processedText = cleanUpMessage(text, isImpersonate, isContinue, !isFinal, !isFinal);
|
||||
let processedText = cleanUpMessage(text, isImpersonate, isContinue, !isFinal, this.stoppingStrings);
|
||||
|
||||
// Predict unbalanced asterisks / quotes during streaming
|
||||
const charsToBalance = ['*', '"', '```'];
|
||||
@ -2805,6 +2805,12 @@ class StreamingProcessor {
|
||||
scrollLock = false;
|
||||
}
|
||||
|
||||
// Stopping strings are expensive to calculate, especially with macros enabled. To remove stopping strings
|
||||
// when streaming, we cache the result of getStoppingStrings instead of calling it once per token.
|
||||
const isImpersonate = this.type == 'impersonate';
|
||||
const isContinue = this.type == 'continue';
|
||||
this.stoppingStrings = getStoppingStrings(isImpersonate, isContinue);
|
||||
|
||||
try {
|
||||
const sw = new Stopwatch(1000 / power_user.streaming_fps);
|
||||
const timestamps = [];
|
||||
@ -2907,7 +2913,7 @@ export async function generateRaw(prompt, api, instructOverride) {
|
||||
throw new Error(data.error);
|
||||
}
|
||||
|
||||
const message = cleanUpMessage(extractMessageFromData(data), false, false, true, false);
|
||||
const message = cleanUpMessage(extractMessageFromData(data), false, false, true);
|
||||
|
||||
if (!message) {
|
||||
throw new Error('No message generated');
|
||||
@ -3814,7 +3820,7 @@ async function Generate(type, { automatic_trigger, force_name2, quiet_prompt, qu
|
||||
streamingProcessor.generator = streamingGenerator;
|
||||
hideSwipeButtons();
|
||||
let getMessage = await streamingProcessor.generate();
|
||||
let messageChunk = cleanUpMessage(getMessage, isImpersonate, isContinue, false, false);
|
||||
let messageChunk = cleanUpMessage(getMessage, isImpersonate, isContinue, false);
|
||||
|
||||
if (isContinue) {
|
||||
getMessage = continue_mag + getMessage;
|
||||
@ -3849,7 +3855,7 @@ async function Generate(type, { automatic_trigger, force_name2, quiet_prompt, qu
|
||||
|
||||
const swipes = extractMultiSwipes(data, type);
|
||||
|
||||
messageChunk = cleanUpMessage(getMessage, isImpersonate, isContinue, false, false);
|
||||
messageChunk = cleanUpMessage(getMessage, isImpersonate, isContinue, false);
|
||||
|
||||
if (isContinue) {
|
||||
getMessage = continue_mag + getMessage;
|
||||
@ -3857,7 +3863,7 @@ async function Generate(type, { automatic_trigger, force_name2, quiet_prompt, qu
|
||||
|
||||
//Formating
|
||||
const displayIncomplete = type === 'quiet' && !quietToLoud;
|
||||
getMessage = cleanUpMessage(getMessage, isImpersonate, isContinue, displayIncomplete, false);
|
||||
getMessage = cleanUpMessage(getMessage, isImpersonate, isContinue, displayIncomplete);
|
||||
|
||||
if (getMessage.length > 0) {
|
||||
if (isImpersonate) {
|
||||
@ -4487,7 +4493,7 @@ function extractMultiSwipes(data, type) {
|
||||
}
|
||||
|
||||
for (let i = 1; i < data.choices.length; i++) {
|
||||
const text = cleanUpMessage(data.choices[i].text, false, false, false, false);
|
||||
const text = cleanUpMessage(data.choices[i].text, false, false, false);
|
||||
swipes.push(text);
|
||||
}
|
||||
}
|
||||
@ -4495,7 +4501,7 @@ function extractMultiSwipes(data, type) {
|
||||
return swipes;
|
||||
}
|
||||
|
||||
function cleanUpMessage(getMessage, isImpersonate, isContinue, displayIncompleteSentences = false, skipStopStringCleanup = false) {
|
||||
function cleanUpMessage(getMessage, isImpersonate, isContinue, displayIncompleteSentences = false, stoppingStrings = null) {
|
||||
if (!getMessage) {
|
||||
return '';
|
||||
}
|
||||
@ -4510,16 +4516,18 @@ function cleanUpMessage(getMessage, isImpersonate, isContinue, displayIncomplete
|
||||
getMessage = substituteParams(power_user.user_prompt_bias) + getMessage;
|
||||
}
|
||||
|
||||
if (!skipStopStringCleanup) {
|
||||
const stoppingStrings = getStoppingStrings(isImpersonate, isContinue);
|
||||
// Allow for caching of stopping strings. getStoppingStrings is an expensive function, especially with macros
|
||||
// enabled, so for streaming, we call it once and then pass it into each cleanUpMessage call.
|
||||
if (!stoppingStrings) {
|
||||
stoppingStrings = getStoppingStrings(isImpersonate, isContinue);
|
||||
}
|
||||
|
||||
for (const stoppingString of stoppingStrings) {
|
||||
if (stoppingString.length) {
|
||||
for (let j = stoppingString.length; j > 0; j--) {
|
||||
if (getMessage.slice(-j) === stoppingString.slice(0, j)) {
|
||||
getMessage = getMessage.slice(0, -j);
|
||||
break;
|
||||
}
|
||||
for (const stoppingString of stoppingStrings) {
|
||||
if (stoppingString.length) {
|
||||
for (let j = stoppingString.length; j > 0; j--) {
|
||||
if (getMessage.slice(-j) === stoppingString.slice(0, j)) {
|
||||
getMessage = getMessage.slice(0, -j);
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
@ -63,6 +63,7 @@ const TOKENIZER_URLS = {
|
||||
},
|
||||
[tokenizers.API_KOBOLD]: {
|
||||
count: '/api/tokenizers/remote/kobold/count',
|
||||
encode: '/api/tokenizers/remote/kobold/count',
|
||||
},
|
||||
[tokenizers.MISTRAL]: {
|
||||
encode: '/api/tokenizers/mistral/encode',
|
||||
@ -617,6 +618,32 @@ function getTextTokensFromTextgenAPI(str) {
|
||||
return ids;
|
||||
}
|
||||
|
||||
/**
|
||||
* Calls the AI provider's tokenize API to encode a string to tokens.
|
||||
* @param {string} str String to tokenize.
|
||||
* @returns {number[]} Array of token ids.
|
||||
*/
|
||||
function getTextTokensFromKoboldAPI(str) {
|
||||
let ids = [];
|
||||
|
||||
jQuery.ajax({
|
||||
async: false,
|
||||
type: 'POST',
|
||||
url: TOKENIZER_URLS[tokenizers.API_KOBOLD].encode,
|
||||
data: JSON.stringify({
|
||||
text: str,
|
||||
url: api_server,
|
||||
}),
|
||||
dataType: 'json',
|
||||
contentType: 'application/json',
|
||||
success: function (data) {
|
||||
ids = data.ids;
|
||||
},
|
||||
});
|
||||
|
||||
return ids;
|
||||
}
|
||||
|
||||
/**
|
||||
* Calls the underlying tokenizer model to decode token ids to text.
|
||||
* @param {string} endpoint API endpoint.
|
||||
@ -650,6 +677,8 @@ export function getTextTokens(tokenizerType, str) {
|
||||
return getTextTokens(currentRemoteTokenizerAPI(), str);
|
||||
case tokenizers.API_TEXTGENERATIONWEBUI:
|
||||
return getTextTokensFromTextgenAPI(str);
|
||||
case tokenizers.API_KOBOLD:
|
||||
return getTextTokensFromKoboldAPI(str);
|
||||
default: {
|
||||
const tokenizerEndpoints = TOKENIZER_URLS[tokenizerType];
|
||||
if (!tokenizerEndpoints) {
|
||||
|
@ -1438,7 +1438,7 @@ app.use('/api/serpapi', require('./src/endpoints/serpapi').router);
|
||||
// The different text generation APIs
|
||||
|
||||
// Ooba/OpenAI text completions
|
||||
app.use('/api/backends/ooba', require('./src/endpoints/backends/ooba').router);
|
||||
app.use('/api/backends/text-completions', require('./src/endpoints/backends/text-completions').router);
|
||||
|
||||
// KoboldAI
|
||||
app.use('/api/textgen/kobold', require('./src/endpoints/textgen/kobold').router);
|
||||
|
@ -1,4 +1,5 @@
|
||||
const express = require('express');
|
||||
const fetch = require('node-fetch').default;
|
||||
|
||||
const { jsonParser } = require('../../express-common');
|
||||
const { TEXTGEN_TYPES } = require('../../constants');
|
@ -562,7 +562,8 @@ router.post('/remote/kobold/count', jsonParser, async function (request, respons
|
||||
|
||||
const data = await result.json();
|
||||
const count = data['value'];
|
||||
return response.send({ count, ids: [] });
|
||||
const ids = data['ids'] ?? [];
|
||||
return response.send({ count, ids });
|
||||
} catch (error) {
|
||||
console.log(error);
|
||||
return response.send({ error: true });
|
||||
@ -617,7 +618,7 @@ router.post('/remote/textgenerationwebui/encode', jsonParser, async function (re
|
||||
|
||||
const data = await result.json();
|
||||
const count = legacyApi ? data?.results[0]?.tokens : (data?.length ?? data?.value);
|
||||
const ids = legacyApi ? [] : (data?.tokens ?? []);
|
||||
const ids = legacyApi ? [] : (data?.tokens ?? data?.ids ?? []);
|
||||
|
||||
return response.send({ count, ids });
|
||||
} catch (error) {
|
||||
|
@ -349,7 +349,7 @@ function getImages(path) {
|
||||
|
||||
/**
|
||||
* Pipe a fetch() response to an Express.js Response, including status code.
|
||||
* @param {Response} from The Fetch API response to pipe from.
|
||||
* @param {import('node-fetch').Response} from The Fetch API response to pipe from.
|
||||
* @param {Express.Response} to The Express response to pipe to.
|
||||
*/
|
||||
function forwardFetchResponse(from, to) {
|
||||
|
Reference in New Issue
Block a user