Merge pull request #3396 from pcpthm/text-completion-include-reasoning

Support reasoning for OpenRouter Text Completion

public/index.html

@@ -1587,6 +1587,10 @@
                                                 <input type="checkbox" id="skip_special_tokens_textgenerationwebui" />
                                                 <small data-i18n="Skip Special Tokens">Skip Special Tokens</small>
                                             </label>
+                                            <label data-tg-type="openrouter" class="checkbox_label flexGrow flexShrink" for="include_reasoning_textgenerationwebui">
+                                                <input type="checkbox" id="include_reasoning_textgenerationwebui" />
+                                                <small data-i18n="Request Model Reasoning">Request Model Reasoning</small>
+                                            </label>
                                             <label data-tg-type="ooba, aphrodite, tabby" class="checkbox_label  flexGrow flexShrink" for="temperature_last_textgenerationwebui">
                                                 <input type="checkbox" id="temperature_last_textgenerationwebui" />
                                                 <label>

public/script.js

@@ -5706,7 +5706,17 @@ function extractMessageFromData(data) {
  * @returns {string} Extracted reasoning
  */
 function extractReasoningFromData(data) {
-    if (main_api === 'openai' && oai_settings.show_thoughts) {
+    switch (main_api) {
+        case 'textgenerationwebui':
+            switch (textgen_settings.type) {
+                case textgen_types.OPENROUTER:
+                    return data?.choices?.[0]?.reasoning ?? '';
+            }
+            break;
+
+        case 'openai':
+            if (!oai_settings.show_thoughts) break;
+
             switch (oai_settings.chat_completion_source) {
                 case chat_completion_sources.DEEPSEEK:
                     return data?.choices?.[0]?.message?.reasoning_content ?? '';
@@ -5715,6 +5725,7 @@ function extractReasoningFromData(data) {
                 case chat_completion_sources.MAKERSUITE:
                     return data?.responseContent?.parts?.filter(part => part.thought)?.map(part => part.text)?.join('\n\n') ?? '';
             }
+            break;
     }
 
     return '';

public/scripts/openai.js

@@ -298,7 +298,7 @@ const default_settings = {
     names_behavior: character_names_behavior.DEFAULT,
     continue_postfix: continue_postfix_types.SPACE,
     custom_prompt_post_processing: custom_prompt_post_processing_types.NONE,
-    show_thoughts: false,
+    show_thoughts: true,
     seed: -1,
     n: 1,
 };
@@ -377,7 +377,7 @@ const oai_settings = {
     names_behavior: character_names_behavior.DEFAULT,
     continue_postfix: continue_postfix_types.SPACE,
     custom_prompt_post_processing: custom_prompt_post_processing_types.NONE,
-    show_thoughts: false,
+    show_thoughts: true,
     seed: -1,
     n: 1,
 };
@@ -1913,7 +1913,7 @@ async function sendOpenAIRequest(type, messages, signal) {
         'user_name': name1,
         'char_name': name2,
         'group_names': getGroupNames(),
-        'show_thoughts': Boolean(oai_settings.show_thoughts),
+        'include_reasoning': Boolean(oai_settings.show_thoughts),
     };
 
     // Empty array will produce a validation error

public/scripts/preset-manager.js

@@ -586,6 +586,7 @@ class PresetManager {
             'tabby_model',
             'derived',
             'generic_model',
+            'include_reasoning',
         ];
         const settings = Object.assign({}, getSettingsByApiId(this.apiId));
 

public/scripts/textgen-settings.js

@@ -172,6 +172,7 @@ const settings = {
     //truncation_length: 2048,
     ban_eos_token: false,
     skip_special_tokens: true,
+    include_reasoning: true,
     streaming: false,
     mirostat_mode: 0,
     mirostat_tau: 5,
@@ -263,6 +264,7 @@ export const setting_names = [
     'add_bos_token',
     'ban_eos_token',
     'skip_special_tokens',
+    'include_reasoning',
     'streaming',
     'mirostat_mode',
     'mirostat_tau',
@@ -740,6 +742,7 @@ jQuery(function () {
             'add_bos_token_textgenerationwebui': true,
             'temperature_last_textgenerationwebui': true,
             'skip_special_tokens_textgenerationwebui': true,
+            'include_reasoning_textgenerationwebui': true,
             'top_a_textgenerationwebui': 0,
             'top_a_counter_textgenerationwebui': 0,
             'mirostat_mode_textgenerationwebui': 0,
@@ -986,7 +989,7 @@ export async function generateTextGenWithStreaming(generate_data, signal) {
         let logprobs = null;
         const swipes = [];
         const toolCalls = [];
-        const state = {};
+        const state = { reasoning: '' };
         while (true) {
             const { done, value } = await reader.read();
             if (done) return;
@@ -1003,6 +1006,7 @@ export async function generateTextGenWithStreaming(generate_data, signal) {
                 const newText = data?.choices?.[0]?.text || data?.content || '';
                 text += newText;
                 logprobs = parseTextgenLogprobs(newText, data.choices?.[0]?.logprobs || data?.completion_probabilities);
+                state.reasoning += data?.choices?.[0]?.reasoning ?? '';
             }
 
             yield { text, swipes, logprobs, toolCalls, state };
@@ -1266,6 +1270,7 @@ export function getTextGenGenerationData(finalPrompt, maxTokens, isImpersonate,
         'truncation_length': max_context,
         'ban_eos_token': settings.ban_eos_token,
         'skip_special_tokens': settings.skip_special_tokens,
+        'include_reasoning': settings.include_reasoning,
         'top_a': settings.top_a,
         'tfs': settings.tfs,
         'epsilon_cutoff': [OOBA, MANCER].includes(settings.type) ? settings.epsilon_cutoff : undefined,

src/constants.js

@@ -369,6 +369,7 @@ export const OPENROUTER_KEYS = [
     'prompt',
     'stop',
     'provider',
+    'include_reasoning',
 ];
 
 // https://github.com/vllm-project/vllm/blob/0f8a91401c89ac0a8018def3756829611b57727f/vllm/entrypoints/openai/protocol.py#L220

src/endpoints/backends/chat-completions.js

@@ -288,7 +288,7 @@ async function sendMakerSuiteRequest(request, response) {
 
     const model = String(request.body.model);
     const stream = Boolean(request.body.stream);
-    const showThoughts = Boolean(request.body.show_thoughts);
+    const showThoughts = Boolean(request.body.include_reasoning);
     const isThinking = model.includes('thinking');
 
     const generationConfig = {
@@ -998,7 +998,7 @@ router.post('/generate', jsonParser, function (request, response) {
             bodyParams['route'] = 'fallback';
         }
 
-        if (request.body.show_thoughts) {
+        if (request.body.include_reasoning) {
             bodyParams['include_reasoning'] = true;
         }