Add OpenAI o1

package-lock.json

@@ -46,7 +46,7 @@
                 "sanitize-filename": "^1.6.3",
                 "sillytavern-transformers": "2.14.6",
                 "simple-git": "^3.19.1",
-                "tiktoken": "^1.0.15",
+                "tiktoken": "^1.0.16",
                 "vectra": "^0.2.2",
                 "wavefile": "^11.0.0",
                 "write-file-atomic": "^5.0.1",
@@ -5751,9 +5751,10 @@
             "license": "MIT"
         },
         "node_modules/tiktoken": {
-            "version": "1.0.15",
+            "version": "1.0.16",
-            "resolved": "https://registry.npmjs.org/tiktoken/-/tiktoken-1.0.15.tgz",
+            "resolved": "https://registry.npmjs.org/tiktoken/-/tiktoken-1.0.16.tgz",
-            "integrity": "sha512-sCsrq/vMWUSEW29CJLNmPvWxlVp7yh2tlkAjpJltIKqp5CKf98ZNpdeHRmAlPVFlGEbswDc6SmI8vz64W/qErw=="
+            "integrity": "sha512-hRcORIGF2YlAgWx3nzrGJOrKSJwLoc81HpXmMQk89632XAgURc7IeV2FgQ2iXo9z/J96fCvpsHg2kWoHcbj9fg==",
+            "license": "MIT"
         },
         "node_modules/timm": {
             "version": "1.7.1",

package.json

@@ -36,7 +36,7 @@
         "sanitize-filename": "^1.6.3",
         "sillytavern-transformers": "2.14.6",
         "simple-git": "^3.19.1",
-        "tiktoken": "^1.0.15",
+        "tiktoken": "^1.0.16",
         "vectra": "^0.2.2",
         "wavefile": "^11.0.0",
         "write-file-atomic": "^5.0.1",

public/index.html

@@ -383,7 +383,7 @@
                                         Max Response Length (tokens)
                                     </div>
                                     <div class="wide100p">
-                                        <input type="number" id="openai_max_tokens" name="openai_max_tokens" class="text_pole" min="1" max="16384">
+                                        <input type="number" id="openai_max_tokens" name="openai_max_tokens" class="text_pole" min="1" max="65536 ">
                                     </div>
                                 </div>
                                 <div class="range-block" data-source="openai,custom">
@@ -2611,6 +2611,10 @@
                                         <option value="gpt-4-0125-preview">gpt-4-0125-preview (2024)</option>
                                         <option value="gpt-4-1106-preview">gpt-4-1106-preview (2023)</option>
                                     </optgroup>
+                                    <optgroup label="o1">
+                                        <option value="o1-preview">o1-preview</option>
+                                        <option value="o1-mini">o1-mini</option>
+                                    </optgroup>
                                     <optgroup label="Other">
                                         <option value="text-davinci-003">text-davinci-003</option>
                                         <option value="text-davinci-002">text-davinci-002</option>

public/script.js

@@ -2862,7 +2862,12 @@ export function getCharacterCardFields() {
 
 export function isStreamingEnabled() {
     const noStreamSources = [chat_completion_sources.SCALE];
-    return ((main_api == 'openai' && oai_settings.stream_openai && !noStreamSources.includes(oai_settings.chat_completion_source) && !(oai_settings.chat_completion_source == chat_completion_sources.MAKERSUITE && oai_settings.google_model.includes('bison')))
+    return (
+        (main_api == 'openai' &&
+        oai_settings.stream_openai &&
+        !noStreamSources.includes(oai_settings.chat_completion_source) &&
+        !(oai_settings.chat_completion_source == chat_completion_sources.OPENAI && oai_settings.openai_model.startsWith('o1-')) &&
+        !(oai_settings.chat_completion_source == chat_completion_sources.MAKERSUITE && oai_settings.google_model.includes('bison')))
         || (main_api == 'kobold' && kai_settings.streaming_kobold && kai_flags.can_use_streaming)
         || (main_api == 'novel' && nai_settings.streaming_novel)
         || (main_api == 'textgenerationwebui' && textgen_settings.streaming));

public/scripts/openai.js

@@ -1797,7 +1797,7 @@ async function sendOpenAIRequest(type, messages, signal) {
     const isQuiet = type === 'quiet';
     const isImpersonate = type === 'impersonate';
     const isContinue = type === 'continue';
-    const stream = oai_settings.stream_openai && !isQuiet && !isScale && !(isGoogle && oai_settings.google_model.includes('bison'));
+    const stream = oai_settings.stream_openai && !isQuiet && !isScale && !(isGoogle && oai_settings.google_model.includes('bison')) && !(isOAI && oai_settings.openai_model.startsWith('o1-'));
     const useLogprobs = !!power_user.request_token_probabilities;
     const canMultiSwipe = oai_settings.n > 1 && !isContinue && !isImpersonate && !isQuiet && (isOAI || isCustom);
 
@@ -1960,12 +1960,33 @@ async function sendOpenAIRequest(type, messages, signal) {
         generate_data['seed'] = oai_settings.seed;
     }
 
-    await eventSource.emit(event_types.CHAT_COMPLETION_SETTINGS_READY, generate_data);
-
     if (isFunctionCallingSupported() && !stream) {
         await registerFunctionTools(type, generate_data);
     }
 
+    if (isOAI && oai_settings.openai_model.startsWith('o1-')) {
+        generate_data.messages.forEach((msg) => {
+            if (msg.role === 'system') {
+                msg.role = 'user';
+            }
+        });
+        delete generate_data.stream;
+        delete generate_data.logprobs;
+        delete generate_data.top_logprobs;
+        delete generate_data.n;
+        delete generate_data.temperature;
+        delete generate_data.top_p;
+        delete generate_data.frequency_penalty;
+        delete generate_data.presence_penalty;
+        delete generate_data.tools;
+        delete generate_data.tool_choice;
+        // IDK if it supports it and I have no way to test it
+        // delete generate_data.logit_bias;
+        // delete generate_data.stop;
+    }
+
+    await eventSource.emit(event_types.CHAT_COMPLETION_SETTINGS_READY, generate_data);
+
     const generate_url = '/api/backends/chat-completions/generate';
     const response = await fetch(generate_url, {
         method: 'POST',
@@ -3905,6 +3926,9 @@ function getMaxContextOpenAI(value) {
     if (oai_settings.max_context_unlocked) {
         return unlocked_max;
     }
+    else if (value.startsWith('o1-')) {
+        return max_128k;
+    }
     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')) {
         return max_128k;
     }

src/endpoints/tokenizers.js

@@ -350,6 +350,10 @@ function getWebTokenizersChunks(tokenizer, ids) {
  * @returns {string} Tokenizer model to use
  */
 function getTokenizerModel(requestModel) {
+    if (requestModel.includes('o1-preview') || requestModel.includes('o1-mini')) {
+        return 'gpt-4o';
+    }
+
     if (requestModel.includes('gpt-4o')) {
         return 'gpt-4o';
     }