Merge branch 'staging' into small-bookmark-updates

917a994ca84de755238c569a6d2a814f1097d52f

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

Parents: 6257f8b7f9f4a72c0ee33b1d4ff02d481445afb8 5dd1d26350b3a6d98f5f53e52150d2e8c5090820
10 files changed, +95 -84Ignore whitespace
package-lock.json+5 -4
@@ -46,7 +46,7 @@
4646 "sanitize-filename": "^1.6.3",
4747 "sillytavern-transformers": "2.14.6",
4848 "simple-git": "^3.19.1",
4949 "tiktoken": "^1.0.1516",
5050 "vectra": "^0.2.2",
5151 "wavefile": "^11.0.0",
5252 "write-file-atomic": "^5.0.1",
@@ -5751,9 +5751,10 @@
57515751 "license": "MIT"
57525752 },
57535753 "node_modules/tiktoken": {
57545754 "version": "1.0.1516",
57555755 "resolved": "https://registry.npmjs.org/tiktoken/-/tiktoken-1.0.1516.tgz",
57565756 "integrity": "sha512-sCsrq/vMWUSEW29CJLNmPvWxlVp7yh2tlkAjpJltIKqp5CKf98ZNpdeHRmAlPVFlGEbswDc6SmI8vz64WhRcORIGF2YlAgWx3nzrGJOrKSJwLoc81HpXmMQk89632XAgURc7IeV2FgQ2iXo9z/qErwJ96fCvpsHg2kWoHcbj9fg==",
5757+ "license": "MIT"
57575758 },
57585759 "node_modules/timm": {
57595760 "version": "1.7.1",
package.json+1 -1
@@ -36,7 +36,7 @@
3636 "sanitize-filename": "^1.6.3",
3737 "sillytavern-transformers": "2.14.6",
3838 "simple-git": "^3.19.1",
3939 "tiktoken": "^1.0.1516",
4040 "vectra": "^0.2.2",
4141 "wavefile": "^11.0.0",
4242 "write-file-atomic": "^5.0.1",
public/index.html+5 -1
@@ -383,7 +383,7 @@
383383 Max Response Length (tokens)
384384 </div>
385385 <div class="wide100p">
386386 <input type="number" id="openai_max_tokens" name="openai_max_tokens" class="text_pole" min="1" max="1638465536">
387387 </div>
388388 </div>
389389 <div class="range-block" data-source="openai,custom">
@@ -2611,6 +2611,10 @@
26112611 <option value="gpt-4-0125-preview">gpt-4-0125-preview (2024)</option>
26122612 <option value="gpt-4-1106-preview">gpt-4-1106-preview (2023)</option>
26132613 </optgroup>
2614+ <optgroup label="o1">
2615+ <option value="o1-preview">o1-preview</option>
2616+ <option value="o1-mini">o1-mini</option>
2617+ </optgroup>
26142618 <optgroup label="Other">
26152619 <option value="text-davinci-003">text-davinci-003</option>
26162620 <option value="text-davinci-002">text-davinci-002</option>
public/script.js+32 -56
@@ -865,7 +865,6 @@ let abortController;
865865
866866//css
867867var css_send_form_display = $('<div id=send_form></div>').css('display');
868-const MAX_GENERATION_LOOPS = 5;
869868
870869var kobold_horde_model = '';
871870
@@ -2848,7 +2847,12 @@ export function getCharacterCardFields() {
28482847
28492848export function isStreamingEnabled() {
28502849 const noStreamSources = [chat_completion_sources.SCALE];
2851- 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')))
2850+ return (
2851+ (main_api == 'openai' &&
2852+ oai_settings.stream_openai &&
2853+ !noStreamSources.includes(oai_settings.chat_completion_source) &&
2854+ !(oai_settings.chat_completion_source == chat_completion_sources.OPENAI && oai_settings.openai_model.startsWith('o1-')) &&
2855+ !(oai_settings.chat_completion_source == chat_completion_sources.MAKERSUITE && oai_settings.google_model.includes('bison')))
28522856 || (main_api == 'kobold' && kai_settings.streaming_kobold && kai_flags.can_use_streaming)
28532857 || (main_api == 'novel' && nai_settings.streaming_novel)
28542858 || (main_api == 'textgenerationwebui' && textgen_settings.streaming));
@@ -3251,7 +3255,7 @@ export async function generateRaw(prompt, api, instructOverride, quietToLoud, sy
32513255 }
32523256
32533257 if (data.error) {
32543258 throw new Error(data.errorresponse);
32553259 }
32563260
32573261 const message = cleanUpMessage(extractMessageFromData(data), false, false, true);
@@ -3323,11 +3327,11 @@ function removeLastMessage() {
33233327 * @param {GenerateOptions} options Generation options
33243328 * @param {boolean} dryRun Whether to actually generate a message or just assemble the prompt
33253329 * @returns {Promise<any>} Returns a promise that resolves when the text is done generating.
33263330 * @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
33273331 */
33283332export async function Generate(type, { automatic_trigger, force_name2, quiet_prompt, quietToLoud, skipWIAN, force_chid, signal, quietImage, maxLoops, quietName } = {}, dryRun = false) {
33293333 console.log('Generate entered');
33303334 await eventSource.emit(event_types.GENERATION_STARTED, type, { automatic_trigger, force_name2, quiet_prompt, quietToLoud, skipWIAN, force_chid, signal, quietImage, maxLoops }, dryRun);
33313335 setGenerationProgress(0);
33323336 generation_started = new Date();
33333337
@@ -3389,7 +3393,7 @@ export async function Generate(type, { automatic_trigger, force_name2, quiet_pro
33893393 if (selected_group && !is_group_generating) {
33903394 if (!dryRun) {
33913395 // Returns the promise that generateGroupWrapper returns; resolves when generation is done
33923396 return generateGroupWrapper(false, type, { quiet_prompt, force_chid, signal: abortController.signal, quietImage, maxLoops });
33933397 }
33943398
33953399 const characterIndexMap = new Map(characters.map((char, index) => [char.avatar, index]));
@@ -4399,9 +4403,9 @@ export async function Generate(type, { automatic_trigger, force_name2, quiet_pro
43994403 generatedPromptCache = '';
44004404
44014405 if (data?.response) {
44024406 toastr.error(data.response, 'API Error', { preventDuplicates: true });
44034407 }
44044408 throw new Error(data?.response);
44054409 }
44064410
44074411 //const getData = await response.json();
@@ -4421,53 +4425,30 @@ export async function Generate(type, { automatic_trigger, force_name2, quiet_pro
44214425 const displayIncomplete = type === 'quiet' && !quietToLoud;
44224426 getMessage = cleanUpMessage(getMessage, isImpersonate, isContinue, displayIncomplete);
44234427
4424- if (getMessage.length > 0 || data.allowEmptyResponse) {
4428+ if (isImpersonate) {
4425- if (isImpersonate) {
4429+ $('#send_textarea').val(getMessage)[0].dispatchEvent(new Event('input', { bubbles: true }));
4426- $('#send_textarea').val(getMessage)[0].dispatchEvent(new Event('input', { bubbles: true }));
4430+ generatedPromptCache = '';
4427- generatedPromptCache = '';
4431+ await eventSource.emit(event_types.IMPERSONATE_READY, getMessage);
4428- await eventSource.emit(event_types.IMPERSONATE_READY, getMessage);
4432+ }
4429- }
4433+ else if (type == 'quiet') {
4430- else if (type == 'quiet') {
4434+ unblockGeneration(type);
4431- unblockGeneration(type);
4435+ return getMessage;
4432- return getMessage;
4436+ }
4437+ else {
4438+ // Without streaming we'll be having a full message on continuation. Treat it as a last chunk.
4439+ if (originalType !== 'continue') {
4440+ ({ type, getMessage } = await saveReply(type, getMessage, false, title, swipes));
44334441 }
44344442 else {
4435- // Without streaming we'll be having a full message on continuation. Treat it as a last chunk.
4443+ ({ type, getMessage } = await saveReply('appendFinal', getMessage, false, title, swipes));
4436- if (originalType !== 'continue') {
4437- ({ type, getMessage } = await saveReply(type, getMessage, false, title, swipes));
4438- }
4439- else {
4440- ({ type, getMessage } = await saveReply('appendFinal', getMessage, false, title, swipes));
4441- }
4442-
4443- // This relies on `saveReply` having been called to add the message to the chat, so it must be last.
4444- parseAndSaveLogprobs(data, continue_mag);
44454444 }
44464445
4447- if (type !== 'quiet') {
4446+ // This relies on `saveReply` having been called to add the message to the chat, so it must be last.
4448- playMessageSound();
4447+ parseAndSaveLogprobs(data, continue_mag);
44494448 }
4450- } else {
4451- // If maxLoops is not passed in (e.g. first time generating), set it to MAX_GENERATION_LOOPS
4452- maxLoops ??= MAX_GENERATION_LOOPS;
4453-
4454- if (maxLoops === 0) {
4455- if (type !== 'quiet') {
4456- throwCircuitBreakerError();
4457- }
4458- throw new Error('Generate circuit breaker interruption');
4459- }
44604449
4461- // regenerate with character speech reenforced
4450+ if (type !== 'quiet') {
4462- // to make sure we leave on swipe type while also adding the name2 appendage
4451+ playMessageSound();
4463- await delay(1000);
4464- // A message was already deleted on regeneration, so instead treat is as a normal gen
4465- if (type === 'regenerate') {
4466- type = 'normal';
4467- }
4468- // The first await is for waiting for the generate to start. The second one is waiting for it to finish
4469- const result = await await Generate(type, { automatic_trigger, force_name2: true, quiet_prompt, quietToLoud, skipWIAN, force_chid, signal, quietImage, quietName, maxLoops: maxLoops - 1 });
4470- return result;
44714452 }
44724453
44734454 if (power_user.auto_swipe) {
@@ -5240,11 +5221,6 @@ function getGenerateUrl(api) {
52405221 }
52415222}
52425223
5243-function throwCircuitBreakerError() {
5244- callPopup(`Could not extract reply in ${MAX_GENERATION_LOOPS} attempts. Try generating again`, 'text');
5245- unblockGeneration();
5246-}
5247-
52485224function extractTitleFromData(data) {
52495225 if (main_api == 'koboldhorde') {
52505226 return data.workerName;
public/scripts/openai.js+29 -6
@@ -1797,7 +1797,7 @@ async function sendOpenAIRequest(type, messages, signal) {
17971797 const isQuiet = type === 'quiet';
17981798 const isImpersonate = type === 'impersonate';
17991799 const isContinue = type === 'continue';
18001800 const stream = oai_settings.stream_openai && !isQuiet && !isScale && !(isGoogle && oai_settings.google_model.includes('bison')) && !(isOAI && oai_settings.openai_model.startsWith('o1-'));
18011801 const useLogprobs = !!power_user.request_token_probabilities;
18021802 const canMultiSwipe = oai_settings.n > 1 && !isContinue && !isImpersonate && !isQuiet && (isOAI || isCustom);
18031803
@@ -1960,12 +1960,35 @@ async function sendOpenAIRequest(type, messages, signal) {
19601960 generate_data['seed'] = oai_settings.seed;
19611961 }
19621962
1963- await eventSource.emit(event_types.CHAT_COMPLETION_SETTINGS_READY, generate_data);
1964-
19651963 if (isFunctionCallingSupported() && !stream) {
19661964 await registerFunctionTools(type, generate_data);
19671965 }
19681966
1967+ if (isOAI && oai_settings.openai_model.startsWith('o1-')) {
1968+ generate_data.messages.forEach((msg) => {
1969+ if (msg.role === 'system') {
1970+ msg.role = 'user';
1971+ }
1972+ });
1973+ generate_data.max_completion_tokens = generate_data.max_tokens;
1974+ delete generate_data.max_tokens;
1975+ delete generate_data.stream;
1976+ delete generate_data.logprobs;
1977+ delete generate_data.top_logprobs;
1978+ delete generate_data.n;
1979+ delete generate_data.temperature;
1980+ delete generate_data.top_p;
1981+ delete generate_data.frequency_penalty;
1982+ delete generate_data.presence_penalty;
1983+ delete generate_data.tools;
1984+ delete generate_data.tool_choice;
1985+ delete generate_data.stop;
1986+ // It does support logit_bias, but the tokenizer used and its effect is yet unknown.
1987+ // delete generate_data.logit_bias;
1988+ }
1989+
1990+ await eventSource.emit(event_types.CHAT_COMPLETION_SETTINGS_READY, generate_data);
1991+
19691992 const generate_url = '/api/backends/chat-completions/generate';
19701993 const response = await fetch(generate_url, {
19711994 method: 'POST',
@@ -2111,7 +2134,6 @@ async function checkFunctionToolCalls(data) {
21112134 const args = toolCall.function;
21122135 console.log('Function tool call:', toolCall);
21132136 await eventSource.emit(event_types.LLM_FUNCTION_TOOL_CALL, args);
2114- data.allowEmptyResponse = true;
21152137 }
21162138 }
21172139
@@ -2125,7 +2147,6 @@ async function checkFunctionToolCalls(data) {
21252147 /** @type {FunctionToolCall} */
21262148 const args = { name: content.name, arguments: JSON.stringify(content.input) };
21272149 await eventSource.emit(event_types.LLM_FUNCTION_TOOL_CALL, args);
2128- data.allowEmptyResponse = true;
21292150 }
21302151 }
21312152 }
@@ -2140,7 +2161,6 @@ async function checkFunctionToolCalls(data) {
21402161 const args = { name: toolCall.name, arguments: JSON.stringify(toolCall.parameters) };
21412162 console.log('Function tool call:', toolCall);
21422163 await eventSource.emit(event_types.LLM_FUNCTION_TOOL_CALL, args);
2143- data.allowEmptyResponse = true;
21442164 }
21452165 }
21462166}
@@ -3905,6 +3925,9 @@ function getMaxContextOpenAI(value) {
39053925 if (oai_settings.max_context_unlocked) {
39063926 return unlocked_max;
39073927 }
3928+ else if (value.startsWith('o1-')) {
3929+ return max_128k;
3930+ }
39083931 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')) {
39093932 return max_128k;
39103933 }
public/scripts/slash-commands.js+9 -1
@@ -2241,7 +2241,7 @@ function setEphemeralStopStrings(value) {
22412241async function generateRawCallback(args, value) {
22422242 if (!value) {
22432243 console.warn('WARN: No argument provided for /genraw command');
22442244 return '';
22452245 }
22462246
22472247 // Prevent generate recursion
@@ -2260,12 +2260,16 @@ async function generateRawCallback(args, value) {
22602260 setEphemeralStopStrings(resolveVariable(args?.stop));
22612261 const result = await generateRaw(value, '', isFalseBoolean(args?.instruct), quietToLoud, systemPrompt, length);
22622262 return result;
2263+ } catch (err) {
2264+ console.error('Error on /genraw generation', err);
2265+ toastr.error(err.message, 'API Error', { preventDuplicates: true });
22632266 } finally {
22642267 if (lock) {
22652268 activateSendButtons();
22662269 }
22672270 flushEphemeralStoppingStrings();
22682271 }
2272+ return '';
22692273}
22702274
22712275/**
@@ -2291,12 +2295,16 @@ async function generateCallback(args, value) {
22912295 const name = args?.name;
22922296 const result = await generateQuietPrompt(value, quietToLoud, false, '', name, length);
22932297 return result;
2298+ } catch (err) {
2299+ console.error('Error on /gen generation', err);
2300+ toastr.error(err.message, 'API Error', { preventDuplicates: true });
22942301 } finally {
22952302 if (lock) {
22962303 activateSendButtons();
22972304 }
22982305 flushEphemeralStoppingStrings();
22992306 }
2307+ return '';
23002308}
23012309
23022310/**
public/scripts/world-info.js+6 -14
@@ -108,6 +108,7 @@ const KNOWN_DECORATORS = ['@@activate', '@@dont_activate'];
108108 * @property {number} [cooldown] The cooldown of the entry
109109 * @property {number} [delay] The delay of the entry
110110 * @property {string[]} [decorators] Array of decorators for the entry
111+ * @property {number} [hash] The hash of the entry
111112 */
112113
113114/**
@@ -383,12 +384,6 @@ class WorldInfoBuffer {
383384 */
384385class WorldInfoTimedEffects {
385386 /**
386- * Cache for entry hashes. Uses weak map to avoid memory leaks.
387- * @type {WeakMap<WIScanEntry, number>}
388- */
389- #entryHashCache = new WeakMap();
390-
391- /**
392387 * Array of chat messages.
393388 * @type {string[]}
394389 */
@@ -485,13 +480,7 @@ class WorldInfoTimedEffects {
485480 * @returns {number} String hash
486481 */
487482 #getEntryHash(entry) {
488- if (this.#entryHashCache.has(entry)) {
483+ return entry.hash;
489- return this.#entryHashCache.get(entry);
490- }
491-
492- const hash = getStringHash(JSON.stringify(entry));
493- this.#entryHashCache.set(entry, hash);
494- return hash;
495484 }
496485
497486 /**
@@ -3603,10 +3592,13 @@ export async function getSortedEntries() {
36033592 // Chat lore always goes first
36043593 entries = [...chatLore.sort(sortFn), ...entries];
36053594
3606- // Parse decorators
3595+ // Calculate hash and parse decorators. Split maps to preserve old hashes.
36073596 entries = entries.map((entry) => {
36083597 const [decorators, content] = parseDecorators(entry.content || '');
36093598 return { ...entry, decorators, content };
3599+ }).map((entry) => {
3600+ const hash = getStringHash(JSON.stringify(entry));
3601+ return { ...entry, hash };
36103602 });
36113603
36123604 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));
src/endpoints/backends/chat-completions.js+1 -0
@@ -965,6 +965,7 @@ router.post('/generate', jsonParser, function (request, response) {
965965 'model': request.body.model,
966966 'temperature': request.body.temperature,
967967 'max_tokens': request.body.max_tokens,
968+ 'max_completion_tokens': request.body.max_completion_tokens,
968969 'stream': request.body.stream,
969970 'presence_penalty': request.body.presence_penalty,
970971 'frequency_penalty': request.body.frequency_penalty,
src/endpoints/backends/text-completions.js+3 -1
@@ -375,7 +375,9 @@ router.post('/generate', jsonParser, async function (request, response) {
375375 }
376376 }
377377 } catch (error) {
378378 letconst valuestatus = { error: true, status: error?.status, response:?? error?.statusTextcode }?? 'UNKNOWN';
379+ const text = error?.error ?? error?.statusText ?? error?.message ?? 'Unknown error on /generate endpoint';
380+ let value = { error: true, status: status, response: text };
379381 console.log('Endpoint error:', error);
380382
381383 if (!response.headersSent) {
src/endpoints/tokenizers.js+4 -0
@@ -350,6 +350,10 @@ function getWebTokenizersChunks(tokenizer, ids) {
350350 * @returns {string} Tokenizer model to use
351351 */
352352function getTokenizerModel(requestModel) {
353+ if (requestModel.includes('o1-preview') || requestModel.includes('o1-mini')) {
354+ return 'gpt-4o';
355+ }
356+
353357 if (requestModel.includes('gpt-4o')) {
354358 return 'gpt-4o';
355359 }