Merge branch 'staging' into small-bookmark-updates

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;
     }

public/scripts/slash-commands.js

@@ -2241,7 +2241,7 @@ function setEphemeralStopStrings(value) {
 async function generateRawCallback(args, value) {
     if (!value) {
         console.warn('WARN: No argument provided for /genraw command');
-        return;
+        return '';
     }
 
     // Prevent generate recursion
@@ -2260,12 +2260,16 @@ async function generateRawCallback(args, value) {
         setEphemeralStopStrings(resolveVariable(args?.stop));
         const result = await generateRaw(value, '', isFalseBoolean(args?.instruct), quietToLoud, systemPrompt, length);
         return result;
+    } catch (err) {
+        console.error('Error on /genraw generation', err);
+        toastr.error(err.message, 'API Error', { preventDuplicates: true });
     } finally {
         if (lock) {
             activateSendButtons();
         }
         flushEphemeralStoppingStrings();
     }
+    return '';
 }
 
 /**
@@ -2291,12 +2295,16 @@ async function generateCallback(args, value) {
         const name = args?.name;
         const result = await generateQuietPrompt(value, quietToLoud, false, '', name, length);
         return result;
+    } catch (err) {
+        console.error('Error on /gen generation', err);
+        toastr.error(err.message, 'API Error', { preventDuplicates: true });
     } finally {
         if (lock) {
             activateSendButtons();
         }
         flushEphemeralStoppingStrings();
     }
+    return '';
 }
 
 /**

public/scripts/world-info.js

@@ -108,6 +108,7 @@ const KNOWN_DECORATORS = ['@@activate', '@@dont_activate'];
  * @property {number} [cooldown] The cooldown of the entry
  * @property {number} [delay] The delay of the entry
  * @property {string[]} [decorators] Array of decorators for the entry
+ * @property {number} [hash] The hash of the entry
  */
 
 /**
@@ -382,12 +383,6 @@ class WorldInfoBuffer {
  * Represents a timed effects manager for World Info.
  */
 class WorldInfoTimedEffects {
-    /**
-     * Cache for entry hashes. Uses weak map to avoid memory leaks.
-     * @type {WeakMap<WIScanEntry, number>}
-     */
-    #entryHashCache = new WeakMap();
-
     /**
      * Array of chat messages.
      * @type {string[]}
@@ -485,13 +480,7 @@ class WorldInfoTimedEffects {
     * @returns {number} String hash
     */
     #getEntryHash(entry) {
-        if (this.#entryHashCache.has(entry)) {
-            return this.#entryHashCache.get(entry);
-        }
-
-        const hash = getStringHash(JSON.stringify(entry));
-        this.#entryHashCache.set(entry, hash);
-        return hash;
+        return entry.hash;
     }
 
     /**
@@ -3603,10 +3592,13 @@ export async function getSortedEntries() {
         // Chat lore always goes first
         entries = [...chatLore.sort(sortFn), ...entries];
 
-        // Parse decorators
+        // Calculate hash and parse decorators. Split maps to preserve old hashes.
         entries = entries.map((entry) => {
             const [decorators, content] = parseDecorators(entry.content || '');
             return { ...entry, decorators, content };
+        }).map((entry) => {
+            const hash = getStringHash(JSON.stringify(entry));
+            return { ...entry, hash };
         });
 
         console.debug(`[WI] Found ${entries.length} world lore entries. Sorted by strategy`, Object.entries(world_info_insertion_strategy).find((x) => x[1] === world_info_character_strategy));