Merge branch 'staging' into tool-calling

public/index.html

@@ -2954,16 +2954,31 @@
                             </div>
                             <h4 data-i18n="Groq Model">Groq Model</h4>
                             <select id="model_groq_select">
-                                <option value="llama-3.1-405b-reasoning">llama-3.1-405b-reasoning</option>
+                                <optgroup label="Llama 3.2">
-                                <option value="llama-3.1-70b-versatile">llama-3.1-70b-versatile</option>
+                                    <option value="llama-3.2-1b-preview">llama-3.2-1b-preview</option>
-                                <option value="llama-3.1-8b-instant">llama-3.1-8b-instant</option>
+                                    <option value="llama-3.2-3b-preview">llama-3.2-3b-preview</option>
-                                <option value="llama3-groq-70b-8192-tool-use-preview">llama3-groq-70b-8192-tool-use-preview</option>
+                                    <option value="llama-3.2-11b-vision-preview">llama-3.2-11b-vision-preview</option>
-                                <option value="llama3-groq-8b-8192-tool-use-preview">llama3-groq-8b-8192-tool-use-preview</option>
+                                    <option value="llama-3.2-90b-vision-preview">llama-3.2-90b-vision-preview</option>
-                                <option value="llama3-8b-8192">llama3-8b-8192</option>
+                                </optgroup>
-                                <option value="llama3-70b-8192">llama3-70b-8192</option>
+                                <optgroup label="Llama 3.1">
-                                <option value="mixtral-8x7b-32768">mixtral-8x7b-32768</option>
+                                    <option value="llama-3.1-8b-instant">llama-3.1-8b-instant</option>
-                                <option value="gemma-7b-it">gemma-7b-it</option>
+                                    <option value="llama-3.1-70b-versatile">llama-3.1-70b-versatile</option>
-                                <option value="gemma2-9b-it">gemma2-9b-it</option>
+                                    <option value="llama-3.1-405b-reasoning">llama-3.1-405b-reasoning</option>
+                                </optgroup>
+                                <optgroup label="Llama 3">
+                                    <option value="llama3-groq-8b-8192-tool-use-preview">llama3-groq-8b-8192-tool-use-preview</option>
+                                    <option value="llama3-groq-70b-8192-tool-use-preview">llama3-groq-70b-8192-tool-use-preview</option>
+                                    <option value="llama3-8b-8192">llama3-8b-8192</option>
+                                    <option value="llama3-70b-8192">llama3-70b-8192</option>
+                                </optgroup>
+                                <optgroup label="Gemma">
+                                    <option value="gemma-7b-it">gemma-7b-it</option>
+                                    <option value="gemma2-9b-it">gemma2-9b-it</option>
+                                </optgroup>
+                                <optgroup label="Other">
+                                    <option value="mixtral-8x7b-32768">mixtral-8x7b-32768</option>
+                                    <option value="llava-v1.5-7b-4096-preview">llava-v1.5-7b-4096-preview</option>
+                                </optgroup>
                             </select>
                         </div>
                         <div id="perplexity_form" data-source="perplexity">

public/scripts/extensions/attachments/index.js

@@ -1,3 +1,4 @@
+import { event_types, eventSource, saveSettingsDebounced } from '../../../script.js';
 import { deleteAttachment, getDataBankAttachments, getDataBankAttachmentsForSource, getFileAttachment, uploadFileAttachmentToServer } from '../../chats.js';
 import { extension_settings, renderExtensionTemplateAsync } from '../../extensions.js';
 import { SlashCommand } from '../../slash-commands/SlashCommand.js';
@@ -196,7 +197,27 @@ async function enableDataBankAttachment(args, value) {
     return '';
 }
 
+function cleanUpAttachments() {
+    let shouldSaveSettings = false;
+    if (extension_settings.character_attachments) {
+        Object.values(extension_settings.character_attachments).flat().filter(a => a.text).forEach(a => {
+            shouldSaveSettings = true;
+            delete a.text;
+        });
+    }
+    if (Array.isArray(extension_settings.attachments)) {
+        extension_settings.attachments.filter(a => a.text).forEach(a => {
+            shouldSaveSettings = true;
+            delete a.text;
+        });
+    }
+    if (shouldSaveSettings) {
+        saveSettingsDebounced();
+    }
+}
+
 jQuery(async () => {
+    eventSource.on(event_types.APP_READY, cleanUpAttachments);
     const manageButton = await renderExtensionTemplateAsync('attachments', 'manage-button', {});
     const attachButton = await renderExtensionTemplateAsync('attachments', 'attach-button', {});
     $('#data_bank_wand_container').append(manageButton);

public/scripts/extensions/caption/index.js

@@ -403,6 +403,7 @@ jQuery(async function () {
                 (extension_settings.caption.source === 'multimodal' && extension_settings.caption.multimodal_api === 'openai' && (secret_state[SECRET_KEYS.OPENAI] || extension_settings.caption.allow_reverse_proxy)) ||
                 (extension_settings.caption.source === 'multimodal' && extension_settings.caption.multimodal_api === 'openrouter' && secret_state[SECRET_KEYS.OPENROUTER]) ||
                 (extension_settings.caption.source === 'multimodal' && extension_settings.caption.multimodal_api === 'zerooneai' && secret_state[SECRET_KEYS.ZEROONEAI]) ||
+                (extension_settings.caption.source === 'multimodal' && extension_settings.caption.multimodal_api === 'groq' && secret_state[SECRET_KEYS.GROQ]) ||
                 (extension_settings.caption.source === 'multimodal' && extension_settings.caption.multimodal_api === 'mistral' && (secret_state[SECRET_KEYS.MISTRALAI] || extension_settings.caption.allow_reverse_proxy)) ||
                 (extension_settings.caption.source === 'multimodal' && extension_settings.caption.multimodal_api === 'google' && (secret_state[SECRET_KEYS.MAKERSUITE] || extension_settings.caption.allow_reverse_proxy)) ||
                 (extension_settings.caption.source === 'multimodal' && extension_settings.caption.multimodal_api === 'anthropic' && (secret_state[SECRET_KEYS.CLAUDE] || extension_settings.caption.allow_reverse_proxy)) ||

public/scripts/extensions/caption/settings.html

@@ -21,6 +21,7 @@
                         <option value="anthropic">Anthropic</option>
                         <option value="custom" data-i18n="Custom (OpenAI-compatible)">Custom (OpenAI-compatible)</option>
                         <option value="google">Google AI Studio</option>
+                        <option value="groq">Groq</option>
                         <option value="koboldcpp">KoboldCpp</option>
                         <option value="llamacpp">llama.cpp</option>
                         <option value="mistral">MistralAI</option>
@@ -60,6 +61,9 @@
                         <option data-type="google" value="gemini-1.5-pro-exp-0801">gemini-1.5-pro-exp-0801</option>
                         <option data-type="google" value="gemini-1.5-pro-exp-0827">gemini-1.5-pro-exp-0827</option>
                         <option data-type="google" value="gemini-pro-vision">gemini-pro-vision</option>
+                        <option data-type="groq" value="llama-3.2-11b-vision-preview">llama-3.2-11b-vision-preview</option>
+                        <option data-type="groq" value="llama-3.2-90b-vision-preview">llama-3.2-90b-vision-preview</option>
+                        <option data-type="groq" value="llava-v1.5-7b-4096-preview">llava-v1.5-7b-4096-preview</option>
                         <option data-type="openrouter" value="openai/gpt-4-vision-preview">openai/gpt-4-vision-preview</option>
                         <option data-type="openrouter" value="openai/gpt-4o">openai/gpt-4o</option>
                         <option data-type="openrouter" value="openai/gpt-4o-2024-05-13">openai/gpt-4o-2024-05-13</option>

public/scripts/extensions/shared.js

@@ -36,7 +36,7 @@ export async function getMultimodalCaption(base64Img, prompt) {
     const isVllm = extension_settings.caption.multimodal_api === 'vllm';
     const base64Bytes = base64Img.length * 0.75;
     const compressionLimit = 2 * 1024 * 1024;
-    if ((['google', 'openrouter', 'mistral'].includes(extension_settings.caption.multimodal_api) && base64Bytes > compressionLimit) || isOoba || isKoboldCpp) {
+    if ((['google', 'openrouter', 'mistral', 'groq'].includes(extension_settings.caption.multimodal_api) && base64Bytes > compressionLimit) || isOoba || isKoboldCpp) {
         const maxSide = 1024;
         base64Img = await createThumbnail(base64Img, maxSide, maxSide, 'image/jpeg');
     }
@@ -135,6 +135,10 @@ function throwIfInvalidModel(useReverseProxy) {
         throw new Error('01.AI API key is not set.');
     }
 
+    if (extension_settings.caption.multimodal_api === 'groq' && !secret_state[SECRET_KEYS.GROQ]) {
+        throw new Error('Groq API key is not set.');
+    }
+
     if (extension_settings.caption.multimodal_api === 'google' && !secret_state[SECRET_KEYS.MAKERSUITE] && !useReverseProxy) {
         throw new Error('Google AI Studio API key is not set.');
     }

public/scripts/extensions/vectors/index.js

@@ -466,13 +466,13 @@ async function ingestDataBankAttachments(source) {
         }
 
         // Download and process the file
-        file.text = await getFileAttachment(file.url);
+        const fileText = await getFileAttachment(file.url);
         console.log(`Vectors: Retrieved file ${file.name} from Data Bank`);
         // Convert kilobytes to string length
         const thresholdLength = settings.size_threshold_db * 1024;
         // Use chunk size from settings if file is larger than threshold
         const chunkSize = file.size > thresholdLength ? settings.chunk_size_db : -1;
-        await vectorizeFile(file.text, file.name, collectionId, chunkSize, settings.overlap_percent_db);
+        await vectorizeFile(fileText, file.name, collectionId, chunkSize, settings.overlap_percent_db);
     }
 
     return dataBankCollectionIds;

public/scripts/openai.js

@@ -4118,7 +4118,10 @@ async function onModelChange() {
         if (oai_settings.max_context_unlocked) {
             $('#openai_max_context').attr('max', unlocked_max);
         }
-        else if (oai_settings.groq_model.includes('llama-3.1')) {
+        else if (oai_settings.groq_model.includes('llama-3.2') && oai_settings.groq_model.includes('-preview')) {
+            $('#openai_max_context').attr('max', max_8k);
+        }
+        else if (oai_settings.groq_model.includes('llama-3.2') || oai_settings.groq_model.includes('llama-3.1')) {
             $('#openai_max_context').attr('max', max_128k);
         }
         else if (oai_settings.groq_model.includes('llama3-groq')) {

src/endpoints/backends/chat-completions.js

@@ -107,6 +107,7 @@ async function sendClaudeRequest(request, response) {
         }
 
         const requestBody = {
+            /** @type {any} */ system: '',
             messages: convertedPrompt.messages,
             model: request.body.model,
             max_tokens: request.body.max_tokens,
@@ -120,6 +121,8 @@ async function sendClaudeRequest(request, response) {
             requestBody.system = enableSystemPromptCache
                 ? [{ type: 'text', text: convertedPrompt.systemPrompt, cache_control: { type: 'ephemeral' } }]
                 : convertedPrompt.systemPrompt;
+        } else {
+            delete requestBody.system;
         }
         /*
         if (Array.isArray(request.body.tools) && request.body.tools.length > 0) {
@@ -158,12 +161,13 @@ async function sendClaudeRequest(request, response) {
             forwardFetchResponse(generateResponse, response);
         } else {
             if (!generateResponse.ok) {
-                console.log(color.red(`Claude API returned error: ${generateResponse.status} ${generateResponse.statusText}\n${await generateResponse.text()}\n${divider}`));
+                const generateResponseText = await generateResponse.text();
+                console.log(color.red(`Claude API returned error: ${generateResponse.status} ${generateResponse.statusText}\n${generateResponseText}\n${divider}`));
                 return response.status(generateResponse.status).send({ error: true });
             }
 
             const generateResponseJson = await generateResponse.json();
-            const responseText = generateResponseJson.content[0].text;
+            const responseText = generateResponseJson?.content?.[0]?.text || '';
             console.log('Claude response:', generateResponseJson);
 
             // Wrap it back to OAI format + save the original content

src/endpoints/openai.js

@@ -55,6 +55,10 @@ router.post('/caption-image', jsonParser, async (request, response) => {
             key = readSecret(request.user.directories, SECRET_KEYS.MISTRALAI);
         }
 
+        if (request.body.api === 'groq') {
+            key = readSecret(request.user.directories, SECRET_KEYS.GROQ);
+        }
+
         if (!key && !request.body.reverse_proxy && ['custom', 'ooba', 'koboldcpp', 'vllm'].includes(request.body.api) === false) {
             console.log('No key found for API', request.body.api);
             return response.sendStatus(400);
@@ -111,6 +115,10 @@ router.post('/caption-image', jsonParser, async (request, response) => {
             apiUrl = 'https://api.01.ai/v1/chat/completions';
         }
 
+        if (request.body.api === 'groq') {
+            apiUrl = 'https://api.groq.com/openai/v1/chat/completions';
+        }
+
         if (request.body.api === 'mistral') {
             apiUrl = 'https://api.mistral.ai/v1/chat/completions';
         }