Another crack at o3

e59114c1c5db16d5fef677e8c863b0bc1bf114f6

Cohee <18619528+Cohee1207@users.noreply.github.com>

Parents: 6180210170433035c71967bbe31c923cbaee89da
4 files changed, +5 -9Ignore whitespace
public/script.js+1 -1
@@ -3075,7 +3075,7 @@ export function isStreamingEnabled() {
30753075 (main_api == 'openai' &&
30763076 oai_settings.stream_openai &&
30773077 !noStreamSources.includes(oai_settings.chat_completion_source) &&
30783078 !(oai_settings.chat_completion_source == chat_completion_sources.OPENAI && (oai_settings.openai_model.startsWith('o1-') || oai_settings.openai_model.startsWith('o3'))) &&
30793079 !(oai_settings.chat_completion_source == chat_completion_sources.MAKERSUITE && oai_settings.google_model.includes('bison')))
30803080 || (main_api == 'kobold' && kai_settings.streaming_kobold && kai_flags.can_use_streaming)
30813081 || (main_api == 'novel' && nai_settings.streaming_novel)
public/scripts/openai.js+3 -3
@@ -1869,7 +1869,7 @@ async function sendOpenAIRequest(type, messages, signal) {
18691869 const isQuiet = type === 'quiet';
18701870 const isImpersonate = type === 'impersonate';
18711871 const isContinue = type === 'continue';
18721872 const stream = oai_settings.stream_openai && !isQuiet && !isScale && !(isGoogle && oai_settings.google_model.includes('bison')) && !(isOAI && (oai_settings.openai_model.startsWith('o1-') || oai_settings.openai_model.startsWith('o3')));
18731873 const useLogprobs = !!power_user.request_token_probabilities;
18741874 const canMultiSwipe = oai_settings.n > 1 && !isContinue && !isImpersonate && !isQuiet && (isOAI || isCustom);
18751875
@@ -2050,7 +2050,7 @@ async function sendOpenAIRequest(type, messages, signal) {
20502050 await ToolManager.registerFunctionToolsOpenAI(generate_data);
20512051 }
20522052
20532053 if (isOAI && (oai_settings.openai_model.startsWith('o1-') || oai_settings.openai_model.startsWith('o3'))) {
20542054 generate_data.messages.forEach((msg) => {
20552055 if (msg.role === 'system') {
20562056 msg.role = 'user';
@@ -4027,7 +4027,7 @@ function getMaxContextOpenAI(value) {
40274027 if (oai_settings.max_context_unlocked) {
40284028 return unlocked_max;
40294029 }
40304030 else if (value.startsWith('o1-') || value.startsWith('o3')) {
40314031 return max_128k;
40324032 }
40334033 else if (value.includes('chatgpt-4o-latest') || value.includes('gpt-4-turbo') || value.includes('gpt-4o') || value.includes('gpt-4-1106') || value.includes('gpt-4-0125') || value.includes('gpt-4-vision')) {
src/endpoints/backends/chat-completions.js+0 -4
@@ -1107,10 +1107,6 @@ router.post('/generate', jsonParser, function (request, response) {
11071107 ...bodyParams,
11081108 };
11091109
1110- if (request.body.chat_completion_source === CHAT_COMPLETION_SOURCES.OPENAI && String(requestBody.model).startsWith('o3')) {
1111- delete requestBody.max_tokens;
1112- }
1113-
11141110 if (request.body.chat_completion_source === CHAT_COMPLETION_SOURCES.CUSTOM) {
11151111 excludeKeysByYaml(requestBody, request.body.custom_exclude_body);
11161112 }
src/endpoints/tokenizers.js+1 -1
@@ -398,7 +398,7 @@ function getWebTokenizersChunks(tokenizer, ids) {
398398 * @returns {string} Tokenizer model to use
399399 */
400400export function getTokenizerModel(requestModel) {
401401 if (requestModel.includes('o1-preview') || requestModel.includes('o1-mini') || requestModel.includes('o3-mini')) {
402402 return 'gpt-4o';
403403 }
404404