Merge pull request #2842 from SillyTavern/o1

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

@@ -881,7 +881,6 @@ let abortController;
 
 //css
 var css_send_form_display = $('<div id=send_form></div>').css('display');
-const MAX_GENERATION_LOOPS = 5;
 
 var kobold_horde_model = '';
 
@@ -2862,7 +2861,12 @@ 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));
@@ -3337,11 +3341,11 @@ function removeLastMessage() {
  * @param {GenerateOptions} options Generation options
  * @param {boolean} dryRun Whether to actually generate a message or just assemble the prompt
  * @returns {Promise<any>} Returns a promise that resolves when the text is done generating.
- * @typedef {{automatic_trigger?: boolean, force_name2?: boolean, quiet_prompt?: string, quietToLoud?: boolean, skipWIAN?: boolean, force_chid?: number, signal?: AbortSignal, quietImage?: string, maxLoops?: number, quietName?: string }} GenerateOptions
+ * @typedef {{automatic_trigger?: boolean, force_name2?: boolean, quiet_prompt?: string, quietToLoud?: boolean, skipWIAN?: boolean, force_chid?: number, signal?: AbortSignal, quietImage?: string, quietName?: string }} GenerateOptions
  */
-export async function Generate(type, { automatic_trigger, force_name2, quiet_prompt, quietToLoud, skipWIAN, force_chid, signal, quietImage, maxLoops, quietName } = {}, dryRun = false) {
+export async function Generate(type, { automatic_trigger, force_name2, quiet_prompt, quietToLoud, skipWIAN, force_chid, signal, quietImage, quietName } = {}, dryRun = false) {
     console.log('Generate entered');
-    await eventSource.emit(event_types.GENERATION_STARTED, type, { automatic_trigger, force_name2, quiet_prompt, quietToLoud, skipWIAN, force_chid, signal, quietImage, maxLoops }, dryRun);
+    await eventSource.emit(event_types.GENERATION_STARTED, type, { automatic_trigger, force_name2, quiet_prompt, quietToLoud, skipWIAN, force_chid, signal, quietImage }, dryRun);
     setGenerationProgress(0);
     generation_started = new Date();
 
@@ -3403,7 +3407,7 @@ export async function Generate(type, { automatic_trigger, force_name2, quiet_pro
     if (selected_group && !is_group_generating) {
         if (!dryRun) {
             // Returns the promise that generateGroupWrapper returns; resolves when generation is done
-            return generateGroupWrapper(false, type, { quiet_prompt, force_chid, signal: abortController.signal, quietImage, maxLoops });
+            return generateGroupWrapper(false, type, { quiet_prompt, force_chid, signal: abortController.signal, quietImage });
         }
 
         const characterIndexMap = new Map(characters.map((char, index) => [char.avatar, index]));
@@ -4435,53 +4439,30 @@ export async function Generate(type, { automatic_trigger, force_name2, quiet_pro
         const displayIncomplete = type === 'quiet' && !quietToLoud;
         getMessage = cleanUpMessage(getMessage, isImpersonate, isContinue, displayIncomplete);
 
-        if (getMessage.length > 0 || data.allowEmptyResponse) {
+        if (isImpersonate) {
-            if (isImpersonate) {
+            $('#send_textarea').val(getMessage)[0].dispatchEvent(new Event('input', { bubbles: true }));
-                $('#send_textarea').val(getMessage)[0].dispatchEvent(new Event('input', { bubbles: true }));
+            generatedPromptCache = '';
-                generatedPromptCache = '';
+            await eventSource.emit(event_types.IMPERSONATE_READY, getMessage);
-                await eventSource.emit(event_types.IMPERSONATE_READY, getMessage);
+        }
-            }
+        else if (type == 'quiet') {
-            else if (type == 'quiet') {
+            unblockGeneration(type);
-                unblockGeneration(type);
+            return getMessage;
-                return getMessage;
+        }
+        else {
+            // Without streaming we'll be having a full message on continuation. Treat it as a last chunk.
+            if (originalType !== 'continue') {
+                ({ type, getMessage } = await saveReply(type, getMessage, false, title, swipes));
             }
             else {
-                // Without streaming we'll be having a full message on continuation. Treat it as a last chunk.
+                ({ type, getMessage } = await saveReply('appendFinal', getMessage, false, title, swipes));
-                if (originalType !== 'continue') {
-                    ({ type, getMessage } = await saveReply(type, getMessage, false, title, swipes));
-                }
-                else {
-                    ({ type, getMessage } = await saveReply('appendFinal', getMessage, false, title, swipes));
-                }
-
-                // This relies on `saveReply` having been called to add the message to the chat, so it must be last.
-                parseAndSaveLogprobs(data, continue_mag);
             }
 
-            if (type !== 'quiet') {
+            // This relies on `saveReply` having been called to add the message to the chat, so it must be last.
-                playMessageSound();
+            parseAndSaveLogprobs(data, continue_mag);
-            }
+        }
-        } else {
-            // If maxLoops is not passed in (e.g. first time generating), set it to MAX_GENERATION_LOOPS
-            maxLoops ??= MAX_GENERATION_LOOPS;
 
-            if (maxLoops === 0) {
+        if (type !== 'quiet') {
-                if (type !== 'quiet') {
+            playMessageSound();
-                    throwCircuitBreakerError();
-                }
-                throw new Error('Generate circuit breaker interruption');
-            }
-
-            // regenerate with character speech reenforced
-            // to make sure we leave on swipe type while also adding the name2 appendage
-            await delay(1000);
-            // A message was already deleted on regeneration, so instead treat is as a normal gen
-            if (type === 'regenerate') {
-                type = 'normal';
-            }
-            // The first await is for waiting for the generate to start. The second one is waiting for it to finish
-            const result = await await Generate(type, { automatic_trigger, force_name2: true, quiet_prompt, quietToLoud, skipWIAN, force_chid, signal, quietImage, quietName, maxLoops: maxLoops - 1 });
-            return result;
         }
 
         if (power_user.auto_swipe) {
@@ -5254,11 +5235,6 @@ function getGenerateUrl(api) {
     }
 }
 
-function throwCircuitBreakerError() {
-    callPopup(`Could not extract reply in ${MAX_GENERATION_LOOPS} attempts. Try generating again`, 'text');
-    unblockGeneration();
-}
-
 function extractTitleFromData(data) {
     if (main_api == 'koboldhorde') {
         return data.workerName;

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 +1960,35 @@ 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';
+            }
+        });
+        generate_data.max_completion_tokens = generate_data.max_tokens;
+        delete generate_data.max_tokens;
+        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;
+        delete generate_data.stop;
+        // It does support logit_bias, but the tokenizer used and its effect is yet unknown.
+        // delete generate_data.logit_bias;
+    }
+
+    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',
@@ -2111,7 +2134,6 @@ async function checkFunctionToolCalls(data) {
             const args = toolCall.function;
             console.log('Function tool call:', toolCall);
             await eventSource.emit(event_types.LLM_FUNCTION_TOOL_CALL, args);
-            data.allowEmptyResponse = true;
         }
     }
 
@@ -2125,7 +2147,6 @@ async function checkFunctionToolCalls(data) {
                 /** @type {FunctionToolCall} */
                 const args = { name: content.name, arguments: JSON.stringify(content.input) };
                 await eventSource.emit(event_types.LLM_FUNCTION_TOOL_CALL, args);
-                data.allowEmptyResponse = true;
             }
         }
     }
@@ -2140,7 +2161,6 @@ async function checkFunctionToolCalls(data) {
             const args = { name: toolCall.name, arguments: JSON.stringify(toolCall.parameters) };
             console.log('Function tool call:', toolCall);
             await eventSource.emit(event_types.LLM_FUNCTION_TOOL_CALL, args);
-            data.allowEmptyResponse = true;
         }
     }
 }
@@ -3905,6 +3925,9 @@ 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/backends/chat-completions.js

@@ -965,6 +965,7 @@ router.post('/generate', jsonParser, function (request, response) {
         'model': request.body.model,
         'temperature': request.body.temperature,
         'max_tokens': request.body.max_tokens,
+        'max_completion_tokens': request.body.max_completion_tokens,
         'stream': request.body.stream,
         'presence_penalty': request.body.presence_penalty,
         'frequency_penalty': request.body.frequency_penalty,

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';
     }