From 46d5f79fd96a1a30562e4b4c3be0e0fa08adc2b8 Mon Sep 17 00:00:00 2001 From: Cohee <18619528+Cohee1207@users.noreply.github.com> Date: Tue, 18 Mar 2025 21:33:11 +0200 Subject: [PATCH 01/25] OpenRouter: Allow applying prompt post-processing Fixes #3689 --- public/index.html | 16 +++++++++------- public/scripts/openai.js | 2 +- src/endpoints/backends/chat-completions.js | 17 +++++++++-------- 3 files changed, 19 insertions(+), 16 deletions(-) diff --git a/public/index.html b/public/index.html index ec0e088a3..d82d7f833 100644 --- a/public/index.html +++ b/public/index.html @@ -3415,13 +3415,6 @@
-

Prompt Post-Processing

-

@@ -3440,6 +3433,15 @@

+
+

Prompt Post-Processing

+ +
Cancel
diff --git a/public/scripts/openai.js b/public/scripts/openai.js index 80d1e9ed7..36e224bf9 100644 --- a/public/scripts/openai.js +++ b/public/scripts/openai.js @@ -2020,6 +2020,7 @@ async function sendOpenAIRequest(type, messages, signal) { 'reasoning_effort': String(oai_settings.reasoning_effort), 'enable_web_search': Boolean(oai_settings.enable_web_search), 'request_images': Boolean(oai_settings.request_images), + 'custom_prompt_post_processing': oai_settings.custom_prompt_post_processing, }; if (!canMultiSwipe && ToolManager.canPerformToolCalls(type)) { @@ -2100,7 +2101,6 @@ async function sendOpenAIRequest(type, messages, signal) { generate_data['custom_include_body'] = oai_settings.custom_include_body; generate_data['custom_exclude_body'] = oai_settings.custom_exclude_body; generate_data['custom_include_headers'] = oai_settings.custom_include_headers; - generate_data['custom_prompt_post_processing'] = oai_settings.custom_prompt_post_processing; } if (isCohere) { diff --git a/src/endpoints/backends/chat-completions.js b/src/endpoints/backends/chat-completions.js index 4d2eee774..e8d3842ad 100644 --- a/src/endpoints/backends/chat-completions.js +++ b/src/endpoints/backends/chat-completions.js @@ -1121,14 +1121,6 @@ router.post('/generate', function (request, response) { mergeObjectWithYaml(bodyParams, request.body.custom_include_body); mergeObjectWithYaml(headers, request.body.custom_include_headers); - - if (request.body.custom_prompt_post_processing) { - console.info('Applying custom prompt post-processing of type', request.body.custom_prompt_post_processing); - request.body.messages = postProcessPrompt( - request.body.messages, - request.body.custom_prompt_post_processing, - getPromptNames(request)); - } } else if (request.body.chat_completion_source === CHAT_COMPLETION_SOURCES.PERPLEXITY) { apiUrl = API_PERPLEXITY; apiKey = readSecret(request.user.directories, SECRET_KEYS.PERPLEXITY); @@ -1160,6 +1152,15 @@ router.post('/generate', function (request, response) { return response.status(400).send({ error: true }); } + const postProcessTypes = [CHAT_COMPLETION_SOURCES.CUSTOM, CHAT_COMPLETION_SOURCES.OPENROUTER]; + if (postProcessTypes.includes(request.body.chat_completion_source) && request.body.custom_prompt_post_processing) { + console.info('Applying custom prompt post-processing of type', request.body.custom_prompt_post_processing); + request.body.messages = postProcessPrompt( + request.body.messages, + request.body.custom_prompt_post_processing, + getPromptNames(request)); + } + // A few of OpenAIs reasoning models support reasoning effort if ([CHAT_COMPLETION_SOURCES.CUSTOM, CHAT_COMPLETION_SOURCES.OPENAI].includes(request.body.chat_completion_source)) { if (['o1', 'o3-mini', 'o3-mini-2025-01-31'].includes(request.body.model)) { From fcaea46a54fdcda610dbbdd94446ee6b383dddde Mon Sep 17 00:00:00 2001 From: Cohee <18619528+Cohee1207@users.noreply.github.com> Date: Tue, 18 Mar 2025 23:17:26 +0200 Subject: [PATCH 02/25] Apply post-process before setting cache at depth --- src/endpoints/backends/chat-completions.js | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-) diff --git a/src/endpoints/backends/chat-completions.js b/src/endpoints/backends/chat-completions.js index e8d3842ad..a39dcbe2f 100644 --- a/src/endpoints/backends/chat-completions.js +++ b/src/endpoints/backends/chat-completions.js @@ -1048,6 +1048,15 @@ router.post('/generate', function (request, response) { let bodyParams; const isTextCompletion = Boolean(request.body.model && TEXT_COMPLETION_MODELS.includes(request.body.model)) || typeof request.body.messages === 'string'; + const postProcessTypes = [CHAT_COMPLETION_SOURCES.CUSTOM, CHAT_COMPLETION_SOURCES.OPENROUTER]; + if (Array.isArray(request.body.messages) && postProcessTypes.includes(request.body.chat_completion_source) && request.body.custom_prompt_post_processing) { + console.info('Applying custom prompt post-processing of type', request.body.custom_prompt_post_processing); + request.body.messages = postProcessPrompt( + request.body.messages, + request.body.custom_prompt_post_processing, + getPromptNames(request)); + } + if (request.body.chat_completion_source === CHAT_COMPLETION_SOURCES.OPENAI) { apiUrl = new URL(request.body.reverse_proxy || API_OPENAI).toString(); apiKey = request.body.reverse_proxy ? request.body.proxy_password : readSecret(request.user.directories, SECRET_KEYS.OPENAI); @@ -1152,15 +1161,6 @@ router.post('/generate', function (request, response) { return response.status(400).send({ error: true }); } - const postProcessTypes = [CHAT_COMPLETION_SOURCES.CUSTOM, CHAT_COMPLETION_SOURCES.OPENROUTER]; - if (postProcessTypes.includes(request.body.chat_completion_source) && request.body.custom_prompt_post_processing) { - console.info('Applying custom prompt post-processing of type', request.body.custom_prompt_post_processing); - request.body.messages = postProcessPrompt( - request.body.messages, - request.body.custom_prompt_post_processing, - getPromptNames(request)); - } - // A few of OpenAIs reasoning models support reasoning effort if ([CHAT_COMPLETION_SOURCES.CUSTOM, CHAT_COMPLETION_SOURCES.OPENAI].includes(request.body.chat_completion_source)) { if (['o1', 'o3-mini', 'o3-mini-2025-01-31'].includes(request.body.model)) { From 5a6058d319196c5e8326678ea177f17ec2bf4ade Mon Sep 17 00:00:00 2001 From: Wolfsblvt Date: Wed, 19 Mar 2025 03:06:50 +0100 Subject: [PATCH 03/25] Adds sprite-based filtering for expressions - Functionality only available for LLM/webLLM - New toggle to filter expressions on availalbe sprites - `getExpressionsList` filters cached expressions when checked (using sprite folder name/override) - `/expression-list` slash command has "filter" arg to filter list - `/expression-classify` slash command has "filter" arg now, to use filtered list for classification - `getExpressionLabel` uses filtered expressions when LLM/webLLM --- .../scripts/extensions/expressions/index.js | 68 +++++++++++++++---- .../extensions/expressions/settings.html | 6 +- 2 files changed, 58 insertions(+), 16 deletions(-) diff --git a/public/scripts/extensions/expressions/index.js b/public/scripts/extensions/expressions/index.js index aa592cfd7..a94908130 100644 --- a/public/scripts/extensions/expressions/index.js +++ b/public/scripts/extensions/expressions/index.js @@ -4,7 +4,7 @@ import { characters, eventSource, event_types, generateRaw, getRequestHeaders, m import { dragElement, isMobile } from '../../RossAscends-mods.js'; import { getContext, getApiUrl, modules, extension_settings, ModuleWorkerWrapper, doExtrasFetch, renderExtensionTemplateAsync } from '../../extensions.js'; import { loadMovingUIState, performFuzzySearch, power_user } from '../../power-user.js'; -import { onlyUnique, debounce, getCharaFilename, trimToEndSentence, trimToStartSentence, waitUntilCondition, findChar } from '../../utils.js'; +import { onlyUnique, debounce, getCharaFilename, trimToEndSentence, trimToStartSentence, waitUntilCondition, findChar, isTrueBoolean } from '../../utils.js'; import { hideMutedSprites, selected_group } from '../../group-chats.js'; import { isJsonSchemaSupported } from '../../textgen-settings.js'; import { debounce_timeout } from '../../constants.js'; @@ -678,7 +678,7 @@ async function setSpriteFolderCommand(_, folder) { return ''; } -async function classifyCallback(/** @type {{api: string?, prompt: string?}} */ { api = null, prompt = null }, text) { +async function classifyCallback(/** @type {{api: string?, filter: string?, prompt: string?}} */ { api = null, filter = null, prompt = null }, text) { if (!text) { toastr.error('No text provided'); return ''; @@ -689,13 +689,14 @@ async function classifyCallback(/** @type {{api: string?, prompt: string?}} */ { } const expressionApi = EXPRESSION_API[api] || extension_settings.expressions.api; + const filterAvailable = isTrueBoolean(filter); if (!modules.includes('classify') && expressionApi == EXPRESSION_API.extras) { toastr.warning('Text classification is disabled or not available'); return ''; } - const label = await getExpressionLabel(text, expressionApi, { customPrompt: prompt }); + const label = await getExpressionLabel(text, expressionApi, { filterAvailable: filterAvailable, customPrompt: prompt }); console.debug(`Classification result for "${text}": ${label}`); return label; } @@ -988,10 +989,11 @@ function onTextGenSettingsReady(args) { * @param {string} text - The text to classify and retrieve the expression label for. * @param {EXPRESSION_API} [expressionsApi=extension_settings.expressions.api] - The expressions API to use for classification. * @param {object} [options={}] - Optional arguments. + * @param {boolean?} [options.filterAvailable=null] - Whether to filter available expressions. If not specified, uses the extension setting. * @param {string?} [options.customPrompt=null] - The custom prompt to use for classification. * @returns {Promise} - The label of the expression. */ -export async function getExpressionLabel(text, expressionsApi = extension_settings.expressions.api, { customPrompt = null } = {}) { +export async function getExpressionLabel(text, expressionsApi = extension_settings.expressions.api, { filterAvailable = null, customPrompt = null } = {}) { // Return if text is undefined, saving a costly fetch request if ((!modules.includes('classify') && expressionsApi == EXPRESSION_API.extras) || !text) { return extension_settings.expressions.fallback_expression; @@ -1003,6 +1005,11 @@ export async function getExpressionLabel(text, expressionsApi = extension_settin text = sampleClassifyText(text); + filterAvailable ??= extension_settings.expressions.filterAvailable; + if (filterAvailable && ![EXPRESSION_API.llm, EXPRESSION_API.webllm].includes(expressionsApi)) { + console.warn('Filter available is only supported for LLM and WebLLM expressions'); + } + try { switch (expressionsApi) { // Local BERT pipeline @@ -1027,7 +1034,7 @@ export async function getExpressionLabel(text, expressionsApi = extension_settin return extension_settings.expressions.fallback_expression; } - const expressionsList = await getExpressionsList(); + const expressionsList = await getExpressionsList({ filterAvailable: filterAvailable }); const prompt = substituteParamsExtended(customPrompt, { labels: expressionsList }) || await getLlmPrompt(expressionsList); eventSource.once(event_types.TEXT_COMPLETION_SETTINGS_READY, onTextGenSettingsReady); const emotionResponse = await generateRaw(text, main_api, false, false, prompt); @@ -1040,7 +1047,7 @@ export async function getExpressionLabel(text, expressionsApi = extension_settin return extension_settings.expressions.fallback_expression; } - const expressionsList = await getExpressionsList(); + const expressionsList = await getExpressionsList({ filterAvailable: filterAvailable }); const prompt = substituteParamsExtended(customPrompt, { labels: expressionsList }) || await getLlmPrompt(expressionsList); const messages = [ { role: 'user', content: text + '\n\n' + prompt }, @@ -1320,12 +1327,28 @@ function getCachedExpressions() { return [...expressionsList, ...extension_settings.expressions.custom].filter(onlyUnique); } -export async function getExpressionsList() { - // Return cached list if available - if (Array.isArray(expressionsList)) { - return getCachedExpressions(); +export async function getExpressionsList({ filterAvailable = false } = {}) { + // If there is no cached list, load and cache it + if (!Array.isArray(expressionsList)) { + expressionsList = await resolveExpressionsList(); } + const expressions = getCachedExpressions(); + + // Filtering is only available for llm and webllm APIs + if (!filterAvailable || ![EXPRESSION_API.llm, EXPRESSION_API.webllm].includes(extension_settings.expressions.api)) { + return expressions; + } + + // Get expressions with available sprites + const currentLastMessage = selected_group ? getLastCharacterMessage() : null; + const spriteFolderName = getSpriteFolderName(currentLastMessage, currentLastMessage?.name); + + return expressions.filter(label => { + const expression = spriteCache[spriteFolderName]?.find(x => x.label === label); + return (expression?.files.length ?? 0) > 0; + }); + /** * Returns the list of expressions from the API or fallback in offline mode. * @returns {Promise} @@ -1372,9 +1395,6 @@ export async function getExpressionsList() { expressionsList = DEFAULT_EXPRESSIONS.slice(); return expressionsList; } - - const result = await resolveExpressionsList(); - return [...result, ...extension_settings.expressions.custom].filter(onlyUnique); } /** @@ -2102,6 +2122,10 @@ function migrateSettings() { extension_settings.expressions.rerollIfSame = !!$(this).prop('checked'); saveSettingsDebounced(); }); + $('#expressions_filter_available').prop('checked', extension_settings.expressions.filterAvailable).on('input', function () { + extension_settings.expressions.filterAvailable = !!$(this).prop('checked'); + saveSettingsDebounced(); + }); $('#expression_override_cleanup_button').on('click', onClickExpressionOverrideRemoveAllButton); $(document).on('dragstart', '.expression', (e) => { e.preventDefault(); @@ -2279,13 +2303,13 @@ function migrateSettings() { SlashCommandParser.addCommandObject(SlashCommand.fromProps({ name: 'expression-list', aliases: ['expressions'], - /** @type {(args: {return: string}) => Promise} */ + /** @type {(args: {return: string, filterAvailable: string}) => Promise} */ callback: async (args) => { let returnType = /** @type {import('../../slash-commands/SlashCommandReturnHelper.js').SlashCommandReturnType} */ (args.return); - const list = await getExpressionsList(); + const list = await getExpressionsList({ filterAvailable: isTrueBoolean(args.filterAvailable) }); return await slashCommandReturnHelper.doReturn(returnType ?? 'pipe', list, { objectToStringFunc: list => list.join(', ') }); }, @@ -2298,6 +2322,13 @@ function migrateSettings() { enumList: slashCommandReturnHelper.enumList({ allowObject: true }), forceEnum: true, }), + SlashCommandNamedArgument.fromProps({ + name: 'filter', + description: 'Filter the list to only include expressions that have available sprites for the current character.', + typeList: [ARGUMENT_TYPE.BOOLEAN], + enumList: commonEnumProviders.boolean('trueFalse')(), + defaultValue: 'true', + }), ], returns: 'The comma-separated list of available expressions, including custom expressions.', helpString: 'Returns a list of available expressions, including custom expressions.', @@ -2313,6 +2344,13 @@ function migrateSettings() { typeList: [ARGUMENT_TYPE.STRING], enumList: Object.keys(EXPRESSION_API).map(api => new SlashCommandEnumValue(api, null, enumTypes.enum)), }), + SlashCommandNamedArgument.fromProps({ + name: 'filter', + description: 'Filter the list to only include expressions that have available sprites for the current character.', + typeList: [ARGUMENT_TYPE.BOOLEAN], + enumList: commonEnumProviders.boolean('trueFalse')(), + defaultValue: 'true', + }), SlashCommandNamedArgument.fromProps({ name: 'prompt', description: 'Custom prompt for classification. Only relevant if Classifier API is set to LLM.', diff --git a/public/scripts/extensions/expressions/settings.html b/public/scripts/extensions/expressions/settings.html index 06e407d7d..98938cf4d 100644 --- a/public/scripts/extensions/expressions/settings.html +++ b/public/scripts/extensions/expressions/settings.html @@ -29,7 +29,11 @@
-