Go team dropdown

bee3cee7407df75a3a1ed80d3ca69cbeb5d33a1e

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

Parents: a95056db40aebef80a5258ca3df5833a39103e6e
5 files changed, +117 -105Ignore whitespace
public/css/toggle-dependent.css+0 -4
@@ -498,7 +498,3 @@ label[for="trim_spaces"]:not(:has(input:checked)) small {
498498#banned_tokens_block_ooba:not(:has(#send_banned_tokens_textgenerationwebui:checked)) #banned_tokens_controls_ooba {
499499 filter: brightness(0.5);
500500}
501-
502-#thinking_budget_controls:not(:has(#enable_thinking:checked)) .range-block:has(#thinking_budget) {
503- filter: brightness(0.5);
504-}
public/index.html+6 -29
@@ -2034,32 +2034,6 @@
20342034 </span>
20352035 </div>
20362036 </div>
2037- <div id="thinking_budget_container" data-source="makersuite">
2038- <div id="thinking_budget_controls">
2039- <div class="range-block">
2040- <label for="enable_thinking" class="checkbox_label widthFreeExpand">
2041- <input id="enable_thinking" type="checkbox" />
2042- <span data-i18n="Enable thinking">Enable thinking</span>
2043- </label>
2044- <div class="toggle-description justifyLeft marginBot5" data-i18n="thinking_budget_desc">
2045- Must be enabled for Thinking Budget to take effect.
2046- </div>
2047- </div>
2048- <div class="range-block">
2049- <div class="range-block-title" data-i18n="Thinking Budget (Tokens)">
2050- Thinking Budget (tokens)
2051- </div>
2052- <div class="range-block-range-and-counter">
2053- <div class="range-block-range">
2054- <input id="thinking_budget" type="range" min="0" max="24576" step="1">
2055- </div>
2056- <div class="range-block-counter">
2057- <input id="thinking_budget_counter" type="number" min="0" max="24576" step="1" data-for="thinking_budget">
2058- </div>
2059- </div>
2060- </div>
2061- </div>
2062- </div>
20632037 <div class="range-block" data-source="deepseek,openrouter,custom,claude,xai">
20642038 <label for="openai_show_thoughts" class="checkbox_label widthFreeExpand">
20652039 <input id="openai_show_thoughts" type="checkbox" />
@@ -2074,16 +2048,19 @@
20742048 </span>
20752049 </div>
20762050 </div>
20772051 <div class="flex-container flexFlowColumn wide100p textAlignCenter marginTop10" data-source="openai,custom,claude,xai,makersuite">
20782052 <div class="flex-container oneline-dropdown" title="Constrains effort on reasoning for reasoning models.&#10;Currently supported values are low, medium, and high.&#10;Reducing reasoning effort can result in faster responses and fewer tokens used on reasoning in a response." data-i18n="[title]Constrains effort on reasoning for reasoning models.">
20792053 <label for="openai_reasoning_effort">
20802054 <span data-i18n="Reasoning Effort">Reasoning Effort</span>
20812055 <i data-source="claude,makersuite" class="opacity50p fa-solid fa-circle-info" title="Allocates a portion of the response length for thinking (low: 10%, medium: 25%, high: 50%), but minimum 1024 tokens."></i>
20822056 </label>
20832057 <select id="openai_reasoning_effort">
2058+ <option data-i18n="openai_reasoning_effort_auto" value="auto">Unset (model-default)</option>
2059+ <option data-i18n="openai_reasoning_effort_minimum" value="min">Mininum (model-defined)</option>
20842060 <option data-i18n="openai_reasoning_effort_low" value="low">Low</option>
20852061 <option data-i18n="openai_reasoning_effort_medium" value="medium">Medium</option>
20862062 <option data-i18n="openai_reasoning_effort_high" value="high">High</option>
2063+ <option data-i18n="openai_reasoning_effort_maximum" value="max">Maximum (model-defined)</option>
20872064 </select>
20882065 </div>
20892066 </div>
public/scripts/openai.js+41 -61
@@ -165,11 +165,6 @@ const textCompletionModels = [
165165 'code-search-ada-code-001',
166166];
167167
168-// One more models list to maintain, yay
169-const thinkingBudgetModels = [
170- 'gemini-2.5-flash-preview-04-17',
171-];
172-
173168let biasCache = undefined;
174169export let model_list = [];
175170
@@ -221,6 +216,15 @@ const openrouter_middleout_types = {
221216 OFF: 'off',
222217};
223218
219+export const reasoning_effort_types = {
220+ auto: 'auto',
221+ low: 'low',
222+ medium: 'medium',
223+ high: 'high',
224+ min: 'min',
225+ max: 'max',
226+};
227+
224228const sensitiveFields = [
225229 'reverse_proxy',
226230 'proxy_password',
@@ -311,8 +315,6 @@ export const settingsToUpdate = {
311315 n: ['#n_openai', 'n', false],
312316 bypass_status_check: ['#openai_bypass_status_check', 'bypass_status_check', true],
313317 request_images: ['#openai_request_images', 'request_images', true],
314- enable_thinking: ['#enable_thinking', 'enable_thinking', true],
315- thinking_budget: ['#thinking_budget', 'thinking_budget', false],
316318};
317319
318320const default_settings = {
@@ -389,11 +391,9 @@ const default_settings = {
389391 continue_postfix: continue_postfix_types.SPACE,
390392 custom_prompt_post_processing: custom_prompt_post_processing_types.NONE,
391393 show_thoughts: true,
392394 reasoning_effort: 'reasoning_effort_types.medium',
393395 enable_web_search: false,
394396 request_images: false,
395- enable_thinking: true,
396- thinking_budget: 1000,
397397 seed: -1,
398398 n: 1,
399399};
@@ -472,11 +472,9 @@ const oai_settings = {
472472 continue_postfix: continue_postfix_types.SPACE,
473473 custom_prompt_post_processing: custom_prompt_post_processing_types.NONE,
474474 show_thoughts: true,
475475 reasoning_effort: 'reasoning_effort_types.medium',
476476 enable_web_search: false,
477477 request_images: false,
478- enable_thinking: true,
479- thinking_budget: 1000,
480478 seed: -1,
481479 n: 1,
482480};
@@ -1948,6 +1946,35 @@ async function sendAltScaleRequest(messages, logit_bias, signal, type) {
19481946 return data.output;
19491947}
19501948
1949+function getReasoningEffort() {
1950+ // Do not set the field. Let the model decide.
1951+ if (oai_settings.reasoning_effort === reasoning_effort_types.auto) {
1952+ return undefined;
1953+ }
1954+
1955+ // These sources require effort as a string
1956+ if (oai_settings.reasoning_effort === reasoning_effort_types.min) {
1957+ switch (oai_settings.chat_completion_source) {
1958+ case chat_completion_sources.OPENAI:
1959+ case chat_completion_sources.CUSTOM:
1960+ case chat_completion_sources.XAI:
1961+ return reasoning_effort_types.low;
1962+ }
1963+ }
1964+
1965+ // Same here, but max effort
1966+ if (oai_settings.reasoning_effort === reasoning_effort_types.max) {
1967+ switch (oai_settings.chat_completion_source) {
1968+ case chat_completion_sources.OPENAI:
1969+ case chat_completion_sources.CUSTOM:
1970+ case chat_completion_sources.XAI:
1971+ return reasoning_effort_types.high;
1972+ }
1973+ }
1974+
1975+ return oai_settings.reasoning_effort;
1976+}
1977+
19511978/**
19521979 * Send a chat completion request to backend
19531980 * @param {string} type (impersonate, quiet, continue, etc)
@@ -2034,17 +2061,12 @@ async function sendOpenAIRequest(type, messages, signal) {
20342061 'char_name': name2,
20352062 'group_names': getGroupNames(),
20362063 'include_reasoning': Boolean(oai_settings.show_thoughts),
20372064 'reasoning_effort': StringgetReasoningEffort(oai_settings.reasoning_effort),
20382065 'enable_web_search': Boolean(oai_settings.enable_web_search),
20392066 'request_images': Boolean(oai_settings.request_images),
20402067 'custom_prompt_post_processing': oai_settings.custom_prompt_post_processing,
20412068 };
20422069
2043- if (thinkingBudgetModels.includes(model)) {
2044- generate_data['enable_thinking'] = oai_settings.enable_thinking;
2045- generate_data['thinking_budget'] = oai_settings.thinking_budget;
2046- }
2047-
20482070 if (!canMultiSwipe && ToolManager.canPerformToolCalls(type)) {
20492071 await ToolManager.registerFunctionToolsOpenAI(generate_data);
20502072 }
@@ -3300,8 +3322,6 @@ function loadOpenAISettings(data, settings) {
33003322 oai_settings.show_thoughts = settings.show_thoughts ?? default_settings.show_thoughts;
33013323 oai_settings.reasoning_effort = settings.reasoning_effort ?? default_settings.reasoning_effort;
33023324 oai_settings.enable_web_search = settings.enable_web_search ?? default_settings.enable_web_search;
3303- oai_settings.enable_thinking = settings.enable_thinking ?? default_settings.enable_thinking;
3304- oai_settings.thinking_budget = settings.thinking_budget ?? default_settings.thinking_budget;
33053325 oai_settings.request_images = settings.request_images ?? default_settings.request_images;
33063326 oai_settings.seed = settings.seed ?? default_settings.seed;
33073327 oai_settings.n = settings.n ?? default_settings.n;
@@ -3437,10 +3457,6 @@ function loadOpenAISettings(data, settings) {
34373457 $('#openai_reasoning_effort').val(oai_settings.reasoning_effort);
34383458 $(`#openai_reasoning_effort option[value="${oai_settings.reasoning_effort}"]`).prop('selected', true);
34393459
3440- $('#enable_thinking').prop('checked', oai_settings.enable_thinking);
3441- $('#thinking_budget').val(oai_settings.thinking_budget);
3442- $('#thinking_budget_counter').val(oai_settings.thinking_budget);
3443-
34443460 if (settings.reverse_proxy !== undefined) oai_settings.reverse_proxy = settings.reverse_proxy;
34453461 $('#openai_reverse_proxy').val(oai_settings.reverse_proxy);
34463462
@@ -3471,7 +3487,6 @@ function loadOpenAISettings(data, settings) {
34713487
34723488 setNamesBehaviorControls();
34733489 setContinuePostfixControls();
3474- updateThinkingBudgetUI();
34753490
34763491 if (oai_settings.custom_prompt_post_processing === custom_prompt_post_processing_types.CLAUDE) {
34773492 oai_settings.custom_prompt_post_processing = custom_prompt_post_processing_types.MERGE;
@@ -3529,14 +3544,6 @@ function setContinuePostfixControls() {
35293544 $('#continue_postfix_display').text(checkedItemText);
35303545}
35313546
3532-/**
3533- * Updates the visibility and state of the Thinking Budget controls.
3534- */
3535-function updateThinkingBudgetUI() {
3536- const modelSupportsControl = thinkingBudgetModels.includes(getChatCompletionModel());
3537- $('#thinking_budget_controls').toggle(modelSupportsControl);
3538-}
3539-
35403547async function getStatusOpen() {
35413548 if (oai_settings.chat_completion_source == chat_completion_sources.WINDOWAI) {
35423549 let status;
@@ -3717,8 +3724,6 @@ async function saveOpenAIPreset(name, settings, triggerUi = true) {
37173724 reasoning_effort: settings.reasoning_effort,
37183725 enable_web_search: settings.enable_web_search,
37193726 request_images: settings.request_images,
3720- enable_thinking: settings.enable_thinking,
3721- thinking_budget: settings.thinking_budget,
37223727 seed: settings.seed,
37233728 n: settings.n,
37243729 };
@@ -4749,7 +4754,6 @@ async function onModelChange() {
47494754
47504755 $('#openai_max_context_counter').attr('max', Number($('#openai_max_context').attr('max')));
47514756
4752- updateThinkingBudgetUI();
47534757 saveSettingsDebounced();
47544758 eventSource.emit(event_types.CHATCOMPLETION_MODEL_CHANGED, value);
47554759}
@@ -5568,7 +5572,6 @@ export function initOpenAI() {
55685572 $('#chat_completion_source').on('change', function () {
55695573 oai_settings.chat_completion_source = String($(this).find(':selected').val());
55705574 toggleChatCompletionForms();
5571- updateThinkingBudgetUI();
55725575 saveSettingsDebounced();
55735576 reconnectOpenAi();
55745577 forceCharacterEditorTokenize();
@@ -5754,29 +5757,6 @@ export function initOpenAI() {
57545757 saveSettingsDebounced();
57555758 });
57565759
5757- $('#enable_thinking').on('input', function () {
5758- oai_settings.enable_thinking = !!$(this).prop('checked');
5759- updateThinkingBudgetUI();
5760- saveSettingsDebounced();
5761- });
5762-
5763- $('#thinking_budget').on('input', function () {
5764- oai_settings.thinking_budget = Number($(this).val());
5765- $('#thinking_budget_counter').val(oai_settings.thinking_budget);
5766- saveSettingsDebounced();
5767- });
5768-
5769- $('#thinking_budget_counter').on('input', function () {
5770- let value = Number($(this).val());
5771- const min = Number($('#thinking_budget').attr('min'));
5772- const max = Number($('#thinking_budget').attr('max'));
5773- value = Math.max(min, Math.min(max, value));
5774- $(this).val(value);
5775- oai_settings.thinking_budget = value;
5776- $('#thinking_budget').val(value);
5777- saveSettingsDebounced();
5778- });
5779-
57805760 $('#openai_enable_web_search').on('input', function () {
57815761 oai_settings.enable_web_search = !!$(this).prop('checked');
57825762 calculateOpenRouterCost();
src/endpoints/backends/chat-completions.js+12 -6
@@ -28,7 +28,8 @@ import {
2828 cachingAtDepthForOpenRouterClaude,
2929 cachingAtDepthForClaude,
3030 getPromptNames,
3131 calculateBudgetTokenscalculateClaudeBudgetTokens,
32+ calculateGoogleBudgetTokens,
3233} from '../../prompt-converters.js';
3334
3435import { readSecret, SECRET_KEYS } from '../secrets.js';
@@ -202,7 +203,7 @@ async function sendClaudeRequest(request, response) {
202203 // No prefill when thinking
203204 voidPrefill = true;
204205 const reasoningEffort = request.body.reasoning_effort;
205206 const budgetTokens = calculateBudgetTokenscalculateClaudeBudgetTokens(requestBody.max_tokens, reasoningEffort, requestBody.stream);
206207 const minThinkTokens = 1024;
207208 if (requestBody.max_tokens <= minThinkTokens) {
208209 const newValue = requestBody.max_tokens + minThinkTokens;
@@ -340,6 +341,7 @@ async function sendMakerSuiteRequest(request, response) {
340341 const stream = Boolean(request.body.stream);
341342 const enableWebSearch = Boolean(request.body.enable_web_search);
342343 const requestImages = Boolean(request.body.request_images);
344+ const reasoningEffort = String(request.body.reasoning_effort);
343345 const isThinking = model.includes('thinking');
344346 const isGemma = model.includes('gemma');
345347
@@ -412,11 +414,15 @@ async function sendMakerSuiteRequest(request, response) {
412414 tools.push({ function_declarations: functionDeclarations });
413415 }
414416
415- if ('enable_thinking' in request.body && 'thinking_budget' in request.body) {
417+ // One more models list to maintain, yay
416- const thinkingEnabled = Boolean(request.body.enable_thinking);
418+ const thinkingBudgetModels = [
417- const thinkingBudget = Number(request.body.thinking_budget);
419+ 'gemini-2.5-flash-preview-04-17',
420+ ];
418421
419422 if (thinkingEnabledthinkingBudgetModels.includes(model)) {
423+ const thinkingBudget = calculateGoogleBudgetTokens(generationConfig.maxOutputTokens, reasoningEffort);
424+
425+ if (Number.isInteger(thinkingBudget)) {
420426 generationConfig.thinkingConfig = { thinkingBudget: thinkingBudget };
421427 }
422428 }
src/prompt-converters.js+58 -5
@@ -3,6 +3,15 @@ import { getConfigValue, tryParse } from './util.js';
33
44const PROMPT_PLACEHOLDER = getConfigValue('promptPlaceholder', 'Let\'s get started.');
55
6+const REASONING_EFFORT = {
7+ auto: 'auto',
8+ low: 'low',
9+ medium: 'medium',
10+ high: 'high',
11+ min: 'min',
12+ max: 'max',
13+};
14+
615/**
716 * @typedef {object} PromptNames
817 * @property {string} charName Character name
@@ -944,25 +953,35 @@ export function cachingAtDepthForOpenRouterClaude(messages, cachingAtDepth) {
944953}
945954
946955/**
947956 * Calculate the Claude budget tokens for a given reasoning effort.
948957 * @param {number} maxTokens Maximum tokens
949958 * @param {string} reasoningEffort Reasoning effort
950959 * @param {boolean} stream If streaming is enabled
951960 * @returns {number} Budget tokens
952961 */
953962export function calculateBudgetTokenscalculateClaudeBudgetTokens(maxTokens, reasoningEffort, stream) {
954963 let budgetTokens = 0;
955964
956965 switch (reasoningEffort) {
957- case 'low':
966+ // Claude doesn't have a default budget value. Use same as min.
967+ case REASONING_EFFORT.auto:
968+ budgetTokens = 1024;
969+ break;
970+ case REASONING_EFFORT.min:
971+ budgetTokens = 1024;
972+ break;
973+ case REASONING_EFFORT.low:
958974 budgetTokens = Math.floor(maxTokens * 0.1);
959975 break;
960976 case 'REASONING_EFFORT.medium':
961977 budgetTokens = Math.floor(maxTokens * 0.25);
962978 break;
963979 case 'REASONING_EFFORT.high':
964980 budgetTokens = Math.floor(maxTokens * 0.5);
965981 break;
982+ case REASONING_EFFORT.max:
983+ budgetTokens = maxTokens;
984+ break;
966985 }
967986
968987 budgetTokens = Math.max(budgetTokens, 1024);
@@ -973,3 +992,37 @@ export function calculateBudgetTokens(maxTokens, reasoningEffort, stream) {
973992
974993 return budgetTokens;
975994}
995+
996+/**
997+ * Calculate the Google budget tokens for a given reasoning effort.
998+ * @param {number} maxTokens Maximum tokens
999+ * @param {string} reasoningEffort Reasoning effort
1000+ * @returns {number?} Budget tokens
1001+ */
1002+export function calculateGoogleBudgetTokens(maxTokens, reasoningEffort) {
1003+ let budgetTokens = 0;
1004+
1005+ switch (reasoningEffort) {
1006+ case REASONING_EFFORT.auto:
1007+ return null;
1008+ case REASONING_EFFORT.min:
1009+ budgetTokens = 0;
1010+ break;
1011+ case REASONING_EFFORT.low:
1012+ budgetTokens = Math.floor(maxTokens * 0.1);
1013+ break;
1014+ case REASONING_EFFORT.medium:
1015+ budgetTokens = Math.floor(maxTokens * 0.25);
1016+ break;
1017+ case REASONING_EFFORT.high:
1018+ budgetTokens = Math.floor(maxTokens * 0.5);
1019+ break;
1020+ case REASONING_EFFORT.max:
1021+ budgetTokens = maxTokens;
1022+ break;
1023+ }
1024+
1025+ budgetTokens = Math.min(budgetTokens, 24576);
1026+
1027+ return budgetTokens;
1028+}