rixty/SillyTavern
1
0
Fork 0
mirror of https://github.com/SillyTavern/SillyTavern.git synced 2025-06-05 21:59:27 +02:00
Code Issues Packages Projects Releases Wiki Activity

Add new OpenAI models

Browse Source
This commit is contained in:
Cohee
2023-06-13 22:33:35 +03:00
parent 96db4a8923
commit 356e85fedd
3 changed files with 39 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
public/index.html
View File

@@ -1442,10 +1442,16 @@
<h4 data-i18n="OpenAI Model">OpenAI Model</h4>
<select id="model_openai_select">
<option value="gpt-3.5-turbo">gpt-3.5-turbo</option>
<option value="gpt-3.5-turbo-16k">gpt-3.5-turbo-16k</option>
<option value="gpt-3.5-turbo-16k-0613">gpt-3.5-turbo-16k-0613</option>
<option value="gpt-3.5-turbo-0613">gpt-3.5-turbo-0613</option>
<option value="gpt-3.5-turbo-0301">gpt-3.5-turbo-0301</option>
<option value="gpt-4">gpt-4</option>
<option value="gpt-4-0613">gpt-4-0613</option>
<option value="gpt-4-0314">gpt-4-0314</option>
<option value="gpt-4-32k">gpt-4-32k</option>
<option value="gpt-4-32k-0301">gpt-4-32k-0301</option>
<option value="gpt-4-32k-0613">gpt-4-32k-0613</option>
</select>
</div>
<div>

13
public/scripts/openai.js
View File

@@ -80,6 +80,7 @@ const default_bias_presets = {
};
const gpt3_max = 4095;
const gpt3_16k_max = 16383;
const gpt4_max = 8191;
const gpt_neox_max = 2048;
const gpt4_32k_max = 32767;
@@ -1540,18 +1541,22 @@ function onModelChange() {
if (oai_settings.max_context_unlocked) {
$('#openai_max_context').attr('max', unlocked_max);
}
else if (value == 'gpt-4' || value == 'gpt-4-0314') {
else if (value == 'gpt-4' || value == 'gpt-4-0314' || value == 'gpt-4-0613') {
$('#openai_max_context').attr('max', gpt4_max);
}
else if (value == 'gpt-4-32k') {
else if (value == 'gpt-4-32k' || value == 'gpt-4-32k-0314' || value == 'gpt-4-32k-0613') {
$('#openai_max_context').attr('max', gpt4_32k_max);
}
else if (value == 'gpt-3.5-turbo-16k' || value == 'gpt-3.5-turbo-16k-0613') {
$('#openai_max_context').attr('max', gpt3_16k_max);
}
else {
$('#openai_max_context').attr('max', gpt3_max);
oai_settings.openai_max_context = Math.max(oai_settings.openai_max_context, gpt3_max);
$('#openai_max_context').val(oai_settings.openai_max_context).trigger('input');
}
oai_settings.openai_max_context = Math.max(oai_settings.openai_max_context, Number($('#openai_max_context').attr('max')));
$('#openai_max_context').val(oai_settings.openai_max_context).trigger('input');
$('#openai_reverse_proxy').attr('placeholder', 'https://api.openai.com/v1');
oai_settings.temp_openai = Math.min(oai_max_temp, oai_settings.temp_openai);

28
server.js
View File

@@ -158,6 +158,23 @@ async function countSentencepieceTokens(spp, text) {
const tokenizersCache = {};
function getTokenizerModel(requestModel) {
if (requestModel.includes('gpt-4-32k')) {
return 'gpt-4-32k';
}
if (requestModel.includes('gpt-4')) {
return 'gpt-4';
}
if (requestModel.includes('gpt-3.5-turbo')) {
return 'gpt-3.5-turbo';
}
// default
return 'gpt-3.5-turbo';
}
function getTiktokenTokenizer(model) {
if (tokenizersCache[model]) {
return tokenizersCache[model];
@@ -2787,7 +2804,8 @@ app.post("/openai_bias", jsonParser, async function (request, response) {
let result = {};
const tokenizer = getTiktokenTokenizer(request.query.model === 'gpt-4-0314' ? 'gpt-4' : request.query.model);
const model = getTokenizerModel(String(request.query.model || ''));
const tokenizer = getTiktokenTokenizer(model);
for (const entry of request.body) {
if (!entry || !entry.text) {
@@ -3083,11 +3101,13 @@ app.post("/generate_openai", jsonParser, function (request, response_generate_op
app.post("/tokenize_openai", jsonParser, function (request, response_tokenize_openai = response) {
if (!request.body) return response_tokenize_openai.sendStatus(400);
const tokensPerName = request.query.model.includes('gpt-4') ? 1 : -1;
const tokensPerMessage = request.query.model.includes('gpt-4') ? 3 : 4;
const model = getTokenizerModel(String(request.query.model || ''));
const tokensPerName = model.includes('gpt-4') ? 1 : -1;
const tokensPerMessage = model.includes('gpt-4') ? 3 : 4;
const tokensPadding = 3;
const tokenizer = getTiktokenTokenizer(request.query.model === 'gpt-4-0314' ? 'gpt-4' : request.query.model);
const tokenizer = getTiktokenTokenizer(model);
let num_tokens = 0;
for (const msg of request.body) {