Merge pull request #3858 from pl752/staging Added option to use secondary API URL in vector extension

8e829c900bdd98be2706c46839195a6c3fbd128b

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

Signed

Parents: 5510e6da31e3798f0262333666bfb91a10fcd4ee 07fb92b37d0fdf793540c1a7bce1d09e3e884240

public/scripts/extensions/vectors/index.js+30 -9 Viewed

 const settings = {
     // For both
     source: 'transformers',
 +    alt_endpoint_url: '',
 +    use_alt_endpoint: false,
     include_wi: false,
     togetherai_model: 'togethercomputer/m2-bert-80M-32k-retrieval',
     openai_model: 'text-embedding-ada-002',

 const moduleWorker = new ModuleWorkerWrapper(synchronizeChat);
 const webllmProvider = new WebLlmVectorProvider();
 const cachedSummaries = new Map();
 +const vectorApiRequiresUrl = ['llamacpp', 'vllm', 'ollama', 'koboldcpp'];
 /**
  * Gets the Collection ID for a file embedded in the chat.

             break;
         case 'ollama':
             body.model = extension_settings.vectors.ollama_model;
             body.apiUrl = settings.use_alt_endpoint ? settings.alt_endpoint_url : textgenerationwebui_settings.server_urls[textgen_types.OLLAMA];
             body.keep = !!extension_settings.vectors.ollama_keep;
             break;
         case 'llamacpp':
             body.apiUrl = settings.use_alt_endpoint ? settings.alt_endpoint_url : textgenerationwebui_settings.server_urls[textgen_types.LLAMACPP];
             break;
         case 'vllm':
             body.apiUrl = settings.use_alt_endpoint ? settings.alt_endpoint_url : textgenerationwebui_settings.server_urls[textgen_types.VLLM];
             body.model = extension_settings.vectors.vllm_model;
             break;
         case 'webllm':

         throw new Error('Vectors: API key missing', { cause: 'api_key_missing' });
+    }
 -    if (settings.source === 'ollama' && !textgenerationwebui_settings.server_urls[textgen_types.OLLAMA] ||
 +    if (vectorApiRequiresUrl.includes(settings.source) && settings.use_alt_endpoint) {
 -        settings.source === 'vllm' && !textgenerationwebui_settings.server_urls[textgen_types.VLLM] ||
 +        if (!settings.alt_endpoint_url) {
 -        settings.source === 'koboldcpp' && !textgenerationwebui_settings.server_urls[textgen_types.KOBOLDCPP] ||
 +            throw new Error('Vectors: API URL missing', { cause: 'api_url_missing' });
 -        settings.source === 'llamacpp' && !textgenerationwebui_settings.server_urls[textgen_types.LLAMACPP]) {
 +        }
 -        throw new Error('Vectors: API URL missing', { cause: 'api_url_missing' });
 +    }
 +    else {
 +        if (settings.source === 'ollama' && !textgenerationwebui_settings.server_urls[textgen_types.OLLAMA] ||
 +            settings.source === 'vllm' && !textgenerationwebui_settings.server_urls[textgen_types.VLLM] ||
 +            settings.source === 'koboldcpp' && !textgenerationwebui_settings.server_urls[textgen_types.KOBOLDCPP] ||
 +            settings.source === 'llamacpp' && !textgenerationwebui_settings.server_urls[textgen_types.LLAMACPP]) {
 +            throw new Error('Vectors: API URL missing', { cause: 'api_url_missing' });
 +        }
+    }
     if (settings.source === 'ollama' && !settings.ollama_model || settings.source === 'vllm' && !settings.vllm_model) {

     $('#webllm_vectorsModel').toggle(settings.source === 'webllm');
     $('#koboldcpp_vectorsModel').toggle(settings.source === 'koboldcpp');
     $('#google_vectorsModel').toggle(settings.source === 'palm');
 +    $('#vector_altEndpointUrl').toggle(vectorApiRequiresUrl.includes(settings.source));
     if (settings.source === 'webllm') {
         loadWebLlmModels();
+    }

         headers: getRequestHeaders(),
         body: JSON.stringify({
             items: items,
             server: settings.use_alt_endpoint ? settings.alt_endpoint_url : textgenerationwebui_settings.server_urls[textgen_types.KOBOLDCPP],
         }),
     });

         saveSettingsDebounced();
         toggleSettings();
     });
 +    $('#vector_altEndpointUrl_enabled').prop('checked', settings.use_alt_endpoint).on('input', () => {
 +        settings.use_alt_endpoint = $('#vector_altEndpointUrl_enabled').prop('checked');
 +        Object.assign(extension_settings.vectors, settings);
 +        saveSettingsDebounced();
 +    });
 +    $('#vector_altEndpoint_address').val(settings.alt_endpoint_url).on('change', () => {
 +        settings.alt_endpoint_url = String($('#vector_altEndpoint_address').val());
 +        Object.assign(extension_settings.vectors, settings);
 +        saveSettingsDebounced();
 +    });
     $('#api_key_nomicai').on('click', async () => {
         const popupText = 'NomicAI API Key:';
         const key = await callGenericPopup(popupText, POPUP_TYPE.INPUT, '', {

public/scripts/extensions/vectors/settings.html+10 -0 Viewed

                     <option value="webllm" data-i18n="WebLLM Extension">WebLLM Extension</option>
                 </select>
             </div>
 +            <div class="flex-container flexFlowColumn" id="vector_altEndpointUrl">
 +                <label class="checkbox_label" for="vector_altEndpointUrl_enabled" title="Enable secondary endpoint URL usage, instead of the main one.">
 +                    <input id="vector_altEndpointUrl_enabled" type="checkbox" class="checkbox">
 +                    <span data-i18n="Use secondary URL">Use secondary URL</span>
 +                </label>
 +                <label for="vector_altEndpoint_address" data-i18n="Secondary Embedding endpoint URL">
 +                    Secondary Embedding endpoint URL
 +                </label>
 +                <input id="vector_altEndpoint_address" class="text_pole" type="text" placeholder="e.g. http://localhost:5001" />
 +            </div>
             <div class="flex-container flexFlowColumn" id="webllm_vectorsModel">
                 <label for="vectors_webllm_model" data-i18n="Vectorization Model">
                     Vectorization Model