Merge branch 'staging' into claude-caching-at-depth

default/content/Char_Avatar_Comfy_Workflow.json

@@ -0,0 +1,137 @@
+{
+  "3": {
+    "inputs": {
+      "seed": "%seed%",
+      "steps": "%steps%",
+      "cfg": "%scale%",
+      "sampler_name": "%sampler%",
+      "scheduler": "%scheduler%",
+      "denoise": "%denoise%",
+      "model": [
+        "4",
+        0
+      ],
+      "positive": [
+        "6",
+        0
+      ],
+      "negative": [
+        "7",
+        0
+      ],
+      "latent_image": [
+        "12",
+        0
+      ]
+    },
+    "class_type": "KSampler",
+    "_meta": {
+      "title": "KSampler"
+    }
+  },
+  "4": {
+    "inputs": {
+      "ckpt_name": "%model%"
+    },
+    "class_type": "CheckpointLoaderSimple",
+    "_meta": {
+      "title": "Load Checkpoint"
+    }
+  },
+  "6": {
+    "inputs": {
+      "text": "%prompt%",
+      "clip": [
+        "4",
+        1
+      ]
+    },
+    "class_type": "CLIPTextEncode",
+    "_meta": {
+      "title": "CLIP Text Encode (Prompt)"
+    }
+  },
+  "7": {
+    "inputs": {
+      "text": "%negative_prompt%",
+      "clip": [
+        "4",
+        1
+      ]
+    },
+    "class_type": "CLIPTextEncode",
+    "_meta": {
+      "title": "CLIP Text Encode (Negative Prompt)"
+    }
+  },
+  "8": {
+    "inputs": {
+      "samples": [
+        "3",
+        0
+      ],
+      "vae": [
+        "4",
+        2
+      ]
+    },
+    "class_type": "VAEDecode",
+    "_meta": {
+      "title": "VAE Decode"
+    }
+  },
+  "9": {
+    "inputs": {
+      "filename_prefix": "SillyTavern",
+      "images": [
+        "8",
+        0
+      ]
+    },
+    "class_type": "SaveImage",
+    "_meta": {
+      "title": "Save Image"
+    }
+  },
+  "10": {
+    "inputs": {
+      "image": "%char_avatar%"
+    },
+    "class_type": "ETN_LoadImageBase64",
+    "_meta": {
+      "title": "Load Image (Base64) [https://github.com/Acly/comfyui-tooling-nodes]"
+    }
+  },
+  "12": {
+    "inputs": {
+      "pixels": [
+        "13",
+        0
+      ],
+      "vae": [
+        "4",
+        2
+      ]
+    },
+    "class_type": "VAEEncode",
+    "_meta": {
+      "title": "VAE Encode"
+    }
+  },
+  "13": {
+    "inputs": {
+      "upscale_method": "bicubic",
+      "width": "%width%",
+      "height": "%height%",
+      "crop": "center",
+      "image": [
+        "10",
+        0
+      ]
+    },
+    "class_type": "ImageScale",
+    "_meta": {
+      "title": "Upscale Image"
+    }
+  }
+}

default/content/index.json

@@ -135,6 +135,10 @@
         "filename": "Default_Comfy_Workflow.json",
         "type": "workflow"
     },
+    {
+        "filename": "Char_Avatar_Comfy_Workflow.json",
+        "type": "workflow"
+    },
     {
         "filename": "presets/kobold/Ace of Spades.json",
         "type": "kobold_preset"

package-lock.json

@@ -3019,9 +3019,9 @@
             }
         },
         "node_modules/cross-spawn": {
-            "version": "7.0.3",
+            "version": "7.0.5",
-            "resolved": "https://registry.npmjs.org/cross-spawn/-/cross-spawn-7.0.3.tgz",
+            "resolved": "https://registry.npmjs.org/cross-spawn/-/cross-spawn-7.0.5.tgz",
-            "integrity": "sha512-iRDPJKUPVEND7dHPO8rkbOnPpyDygcDFtWjpeWNCgy8WP2rXcxXL8TskReQl6OrB2G7+UJrags1q15Fudc7G6w==",
+            "integrity": "sha512-ZVJrKKYunU38/76t0RMOulHOnUcbU9GbpWKAOZ0mhjr7CX6FVrH+4FrAapSOekrgFQ3f/8gwMEuIft0aKq6Hug==",
             "license": "MIT",
             "dependencies": {
                 "path-key": "^3.1.0",

public/css/logprobs.css

@@ -72,6 +72,14 @@
     opacity: 0.5;
 }
 
+.logprobs_output_prefix:hover {
+    background-color: rgba(255, 0, 50, 0.4);
+}
+
+.logprobs_output_prefix:hover ~ .logprobs_output_prefix {
+    background-color: rgba(255, 0, 50, 0.4);
+}
+
 .logprobs_candidate_list {
     grid-row-start: 3;
     grid-row-end: 4;

public/css/scrollable-button.css

@@ -0,0 +1,19 @@
+.scrollable-buttons-container {
+    max-height: 50vh;  /* Use viewport height instead of fixed pixels */
+    overflow-y: auto;
+    -webkit-overflow-scrolling: touch; /* Momentum scrolling on iOS */
+    margin-top: 1rem; /* m-t-1 is equivalent to margin-top: 1rem; */
+    flex-shrink: 1;
+    min-height: 0;
+    scrollbar-width: thin;
+    scrollbar-color: rgba(255, 255, 255, 0.3) transparent;
+}
+
+.scrollable-buttons-container::-webkit-scrollbar {
+    width: 6px;
+}
+
+.scrollable-buttons-container::-webkit-scrollbar-thumb {
+    background-color: rgba(255, 255, 255, 0.3);
+    border-radius: 3px;
+}

public/index.html

@@ -3026,6 +3026,7 @@
                                         <option value="codestral-latest">codestral-latest</option>
                                         <option value="codestral-mamba-latest">codestral-mamba-latest</option>
                                         <option value="pixtral-12b-latest">pixtral-12b-latest</option>
+                                        <option value="pixtral-large-latest">pixtral-large-latest</option>
                                     </optgroup>
                                     <optgroup label="Sub-versions">
                                         <option value="open-mistral-nemo-2407">open-mistral-nemo-2407</option>
@@ -3040,10 +3041,12 @@
                                         <option value="mistral-medium-2312">mistral-medium-2312</option>
                                         <option value="mistral-large-2402">mistral-large-2402</option>
                                         <option value="mistral-large-2407">mistral-large-2407</option>
+                                        <option value="mistral-large-2411">mistral-large-2411</option>
                                         <option value="codestral-2405">codestral-2405</option>
                                         <option value="codestral-2405-blue">codestral-2405-blue</option>
                                         <option value="codestral-mamba-2407">codestral-mamba-2407</option>
                                         <option value="pixtral-12b-2409">pixtral-12b-2409</option>
+                                        <option value="pixtral-large-2411">pixtral-large-2411</option>
                                     </optgroup>
                                     <optgroup id="mistralai_other_models" label="Other"></optgroup>
                                 </select>
@@ -4503,7 +4506,7 @@
                                 </div>
                             </div>
                             <div name="AutoCompleteToggle" class="inline-drawer wide100p flexFlowColumn">
-                                <div class="inline-drawer-toggle inline-drawer-header userSettingsInnerExpandable" title="Options for the various autocompelte input boxes.">
+                                <div class="inline-drawer-toggle inline-drawer-header userSettingsInnerExpandable" title="Options for the various autocomplete input boxes.">
                                     <b><span data-i18n="AutoComplete Settings">AutoComplete Settings</span></b>
                                     <div class="fa-solid fa-circle-chevron-down inline-drawer-icon down"></div>
                                 </div>
@@ -6631,8 +6634,11 @@
                 </div>
             </div>
             <div class="logprobs_panel_content inline-drawer-content flex-container flexFlowColumn">
-                <small>
+                <small class="flex-container alignItemsCenter justifySpaceBetween flexNoWrap">
                     <b data-i18n="Select a token to see alternatives considered by the AI.">Select a token to see alternatives considered by the AI.</b>
+                    <button id="logprobsReroll" class="menu_button margin0" title="Reroll with the entire prefix" data-i18n="[title]Reroll with the entire prefix">
+                        <span class="fa-solid fa-redo logprobs_reroll"></span>
+                    </button>
                 </small>
                 <hr>
                 <div id="logprobs_generation_output"></div>

public/scripts/RossAscends-mods.js

@@ -703,16 +703,13 @@ const isFirefox = navigator.userAgent.toLowerCase().indexOf('firefox') > -1;
  */
 function autoFitSendTextArea() {
     const originalScrollBottom = chatBlock.scrollHeight - (chatBlock.scrollTop + chatBlock.offsetHeight);
-    if (Math.ceil(sendTextArea.scrollHeight + 3) >= Math.floor(sendTextArea.offsetHeight)) {
+
-        const sendTextAreaMinHeight = '0px';
+    sendTextArea.style.height = '1px'; // Reset height to 1px to force recalculation of scrollHeight
-        sendTextArea.style.height = sendTextAreaMinHeight;
+    const newHeight = sendTextArea.scrollHeight;
-    }
-    const newHeight = sendTextArea.scrollHeight + 3;
     sendTextArea.style.height = `${newHeight}px`;
 
     if (!isFirefox) {
-        const newScrollTop = Math.round(chatBlock.scrollHeight - (chatBlock.offsetHeight + originalScrollBottom));
+        chatBlock.scrollTop = chatBlock.scrollHeight - (chatBlock.offsetHeight + originalScrollBottom);
-        chatBlock.scrollTop = newScrollTop;
     }
 }
 export const autoFitSendTextAreaDebounced = debounce(autoFitSendTextArea, debounce_timeout.short);

public/scripts/extensions/caption/settings.html

@@ -37,6 +37,8 @@
                     <select id="caption_multimodal_model" class="flex1 text_pole">
                         <option data-type="mistral" value="pixtral-12b-latest">pixtral-12b-latest</option>
                         <option data-type="mistral" value="pixtral-12b-2409">pixtral-12b-2409</option>
+                        <option data-type="mistral" value="pixtral-large-latest">pixtral-large-latest</option>
+                        <option data-type="mistral" value="pixtral-large-2411">pixtral-large-2411</option>
                         <option data-type="zerooneai" value="yi-vision">yi-vision</option>
                         <option data-type="openai" value="gpt-4-vision-preview">gpt-4-vision-preview</option>
                         <option data-type="openai" value="gpt-4-turbo">gpt-4-turbo</option>

public/scripts/extensions/stable-diffusion/comfyWorkflowEditor.html

@@ -17,6 +17,7 @@
                     <li data-placeholder="scheduler" class="sd_comfy_workflow_editor_not_found">"%scheduler%"</li>
                     <li data-placeholder="steps" class="sd_comfy_workflow_editor_not_found">"%steps%"</li>
                     <li data-placeholder="scale" class="sd_comfy_workflow_editor_not_found">"%scale%"</li>
+                    <li data-placeholder="denoise" class="sd_comfy_workflow_editor_not_found">"%denoise%"</li>
                     <li data-placeholder="clip_skip" class="sd_comfy_workflow_editor_not_found">"%clip_skip%"</li>
                     <li data-placeholder="width" class="sd_comfy_workflow_editor_not_found">"%width%"</li>
                     <li data-placeholder="height" class="sd_comfy_workflow_editor_not_found">"%height%"</li>

public/scripts/extensions/stable-diffusion/index.js

@@ -3269,6 +3269,10 @@ async function generateComfyImage(prompt, negativePrompt, signal) {
 
     const seed = extension_settings.sd.seed >= 0 ? extension_settings.sd.seed : Math.round(Math.random() * Number.MAX_SAFE_INTEGER);
     workflow = workflow.replaceAll('"%seed%"', JSON.stringify(seed));
+
+    const denoising_strength = extension_settings.sd.denoising_strength === undefined ? 1.0 : extension_settings.sd.denoising_strength;
+    workflow = workflow.replaceAll('"%denoise%"', JSON.stringify(denoising_strength));
+
     placeholders.forEach(ph => {
         workflow = workflow.replaceAll(`"%${ph}%"`, JSON.stringify(extension_settings.sd[ph]));
     });
@@ -3279,7 +3283,8 @@ async function generateComfyImage(prompt, negativePrompt, signal) {
         const response = await fetch(getUserAvatarUrl());
         if (response.ok) {
             const avatarBlob = await response.blob();
-            const avatarBase64 = await getBase64Async(avatarBlob);
+            const avatarBase64DataUrl = await getBase64Async(avatarBlob);
+            const avatarBase64 = avatarBase64DataUrl.split(',')[1];
             workflow = workflow.replaceAll('"%user_avatar%"', JSON.stringify(avatarBase64));
         } else {
             workflow = workflow.replaceAll('"%user_avatar%"', JSON.stringify(PNG_PIXEL));
@@ -3289,7 +3294,8 @@ async function generateComfyImage(prompt, negativePrompt, signal) {
         const response = await fetch(getCharacterAvatarUrl());
         if (response.ok) {
             const avatarBlob = await response.blob();
-            const avatarBase64 = await getBase64Async(avatarBlob);
+            const avatarBase64DataUrl = await getBase64Async(avatarBlob);
+            const avatarBase64 = avatarBase64DataUrl.split(',')[1];
             workflow = workflow.replaceAll('"%char_avatar%"', JSON.stringify(avatarBase64));
         } else {
             workflow = workflow.replaceAll('"%char_avatar%"', JSON.stringify(PNG_PIXEL));

public/scripts/extensions/stable-diffusion/settings.html

@@ -319,7 +319,7 @@
                     <input class="neo-range-input" type="number" id="sd_hr_scale_value" data-for="sd_hr_scale" min="{{hr_scale_min}}" max="{{hr_scale_max}}" step="{{hr_scale_step}}" value="{{hr_scale}}" >
                 </div>
 
-                <div class="alignitemscenter flex-container flexFlowColumn flexGrow flexShrink gap0 flexBasis48p" data-sd-source="auto,vlad">
+                <div class="alignitemscenter flex-container flexFlowColumn flexGrow flexShrink gap0 flexBasis48p" data-sd-source="auto,vlad,comfy">
                     <small>
                         <span data-i18n="Denoising strength">Denoising strength</span>
                     </small>

public/scripts/extensions/vectors/index.js

@@ -23,7 +23,7 @@ import {
 import { collapseNewlines, registerDebugFunction } from '../../power-user.js';
 import { SECRET_KEYS, secret_state, writeSecret } from '../../secrets.js';
 import { getDataBankAttachments, getDataBankAttachmentsForSource, getFileAttachment } from '../../chats.js';
-import { debounce, getStringHash as calculateHash, waitUntilCondition, onlyUnique, splitRecursive, trimToStartSentence, trimToEndSentence } from '../../utils.js';
+import { debounce, getStringHash as calculateHash, waitUntilCondition, onlyUnique, splitRecursive, trimToStartSentence, trimToEndSentence, escapeHtml } from '../../utils.js';
 import { debounce_timeout } from '../../constants.js';
 import { getSortedEntries } from '../../world-info.js';
 import { textgen_types, textgenerationwebui_settings } from '../../textgen-settings.js';
@@ -44,6 +44,9 @@ const MODULE_NAME = 'vectors';
 export const EXTENSION_PROMPT_TAG = '3_vectors';
 export const EXTENSION_PROMPT_TAG_DB = '4_vectors_data_bank';
 
+// Force solo chunks for sources that don't support batching.
+const getBatchSize = () => ['transformers', 'palm', 'ollama'].includes(settings.source) ? 1 : 5;
+
 const settings = {
     // For both
     source: 'transformers',
@@ -125,7 +128,7 @@ async function onVectorizeAllClick() {
         // upon request of a full vectorise
         cachedSummaries.clear();
 
-        const batchSize = 5;
+        const batchSize = getBatchSize();
         const elapsedLog = [];
         let finished = false;
         $('#vectorize_progress').show();
@@ -560,7 +563,9 @@ async function vectorizeFile(fileText, fileName, collectionId, chunkSize, overla
             fileText = translatedText;
         }
 
-        const toast = toastr.info('Vectorization may take some time, please wait...', `Ingesting file ${fileName}`);
+        const batchSize = getBatchSize();
+        const toastBody = $('<span>').text('This may take a while. Please wait...');
+        const toast = toastr.info(toastBody, `Ingesting file ${escapeHtml(fileName)}`, { closeButton: false, escapeHtml: false, timeOut: 0, extendedTimeOut: 0 });
         const overlapSize = Math.round(chunkSize * overlapPercent / 100);
         const delimiters = getChunkDelimiters();
         // Overlap should not be included in chunk size. It will be later compensated by overlapChunks
@@ -569,7 +574,12 @@ async function vectorizeFile(fileText, fileName, collectionId, chunkSize, overla
         console.debug(`Vectors: Split file ${fileName} into ${chunks.length} chunks with ${overlapPercent}% overlap`, chunks);
 
         const items = chunks.map((chunk, index) => ({ hash: getStringHash(chunk), text: chunk, index: index }));
-        await insertVectorItems(collectionId, items);
+
+        for (let i = 0; i < items.length; i += batchSize) {
+            toastBody.text(`${i}/${items.length} (${Math.round((i / items.length) * 100)}%) chunks processed`);
+            const chunkedBatch = items.slice(i, i + batchSize);
+            await insertVectorItems(collectionId, chunkedBatch);
+        }
 
         toastr.clear(toast);
         console.log(`Vectors: Inserted ${chunks.length} vector items for file ${fileName} into ${collectionId}`);
@@ -1050,7 +1060,7 @@ async function onViewStatsClick() {
     toastr.info(`Total hashes: <b>${totalHashes}</b><br>
     Unique hashes: <b>${uniqueHashes}</b><br><br>
     I'll mark collected messages with a green circle.`,
-    `Stats for chat ${chatId}`,
+    `Stats for chat ${escapeHtml(chatId)}`,
     { timeOut: 10000, escapeHtml: false },
     );
 

public/scripts/logprobs.js

@@ -1,6 +1,5 @@
 import {
     animation_duration,
-    callPopup,
     chat,
     cleanUpMessage,
     event_types,
@@ -13,15 +12,22 @@ import {
 import { debounce, delay, getStringHash } from './utils.js';
 import { decodeTextTokens, getTokenizerBestMatch } from './tokenizers.js';
 import { power_user } from './power-user.js';
+import { callGenericPopup, POPUP_TYPE } from './popup.js';
+import { t } from './i18n.js';
 
 const TINTS = 4;
 const MAX_MESSAGE_LOGPROBS = 100;
+const REROLL_BUTTON = $('#logprobsReroll');
 
 /**
  * Tuple of a candidate token and its logarithm of probability of being chosen
  * @typedef {[string, number]} Candidate - (token, logprob)
  */
 
+/**
+ * @typedef {(Node|JQuery<Text>|JQuery<HTMLElement>)[]} NodeArray - Array of DOM nodes
+ */
+
 /**
  * Logprob data for a single message
  * @typedef {Object} MessageLogprobData
@@ -43,17 +49,26 @@ const MAX_MESSAGE_LOGPROBS = 100;
  * @property {Candidate[]} topLogprobs - Array of top candidate tokens
  */
 
-let state = {
+/**
-    /** @type {TokenLogprobs | null} */
+ * State object for Token Probabilities
+ * @typedef {Object} LogprobsState
+ * @property {?TokenLogprobs} selectedTokenLogprobs Log probabilities for
+ * currently-selected token.
+ * @property {Map<number, MessageLogprobData>} messageLogprobs Log probabilities for
+ * each message, keyed by message hash.
+ */
+
+/**
+ * @type {LogprobsState} state
+ */
+const state = {
     selectedTokenLogprobs: null,
-    /** @type {Map<number, MessageLogprobData>} */
     messageLogprobs: new Map(),
 };
 
 /**
- * renderAlternativeTokensView renders the Token Probabilities UI and all
+ * Renders the Token Probabilities UI and all subviews with the active message's
- * subviews with the active message's logprobs data. If the message has no token
+ * logprobs data. If the message has no token logprobs, a message is displayed.
- * logprobs, a zero-state is rendered.
  */
 function renderAlternativeTokensView() {
     const view = $('#logprobs_generation_output');
@@ -68,13 +83,14 @@ function renderAlternativeTokensView() {
     const usingSmoothStreaming = isStreamingEnabled() && power_user.smooth_streaming;
     if (!messageLogprobs?.length || usingSmoothStreaming) {
         const emptyState = $('<div></div>');
-        const noTokensMsg = usingSmoothStreaming
+        const noTokensMsg = !power_user.request_token_probabilities
-            ? 'Token probabilities are not available when using Smooth Streaming.'
+            ? '<span>Enable <b>Request token probabilities</b> in the User Settings menu to use this feature.</span>'
-            : 'No token probabilities available for the current message.';
+            : usingSmoothStreaming
-        const msg = power_user.request_token_probabilities
+                ? t`Token probabilities are not available when using Smooth Streaming.`
-            ? noTokensMsg
+                : is_send_press
-            : '<span>Enable <b>Request token probabilities</b> in the User Settings menu to use this feature.</span>';
+                    ? t`Generation in progress...`
-        emptyState.html(msg);
+                    : t`No token probabilities available for the current message.`;
+        emptyState.html(noTokensMsg);
         emptyState.addClass('logprobs_empty_state');
         view.append(emptyState);
         return;
@@ -82,16 +98,34 @@ function renderAlternativeTokensView() {
 
     const prefix = continueFrom || '';
     const tokenSpans = [];
+    REROLL_BUTTON.toggle(!!prefix);
 
     if (prefix) {
-        const prefixSpan = $('<span></span>');
+        REROLL_BUTTON.off('click').on('click', () => onPrefixClicked(prefix.length));
-        prefixSpan.text(prefix);
+
-        prefixSpan.html(prefixSpan.html().replace(/\n/g, '<br>'));
+        let cumulativeOffset = 0;
-        prefixSpan.addClass('logprobs_output_prefix');
+        const words = prefix.split(/\s+/);
-        prefixSpan.attr('title', 'Select to reroll the last \'Continue\' generation.\nHold the CTRL key when clicking to reroll from before that word.');
+        const delimiters = prefix.match(/\s+/g) || []; // Capture the actual delimiters
-        prefixSpan.click(onPrefixClicked);
+
-        addKeyboardProps(prefixSpan);
+        words.forEach((word, i) => {
-        tokenSpans.push(...withVirtualWhitespace(prefix, prefixSpan));
+            const span = $('<span></span>');
+            span.text(`${word} `);
+
+            span.addClass('logprobs_output_prefix');
+            span.attr('title', t`Reroll from this point`);
+
+            let offset = cumulativeOffset;
+            span.on('click', () => onPrefixClicked(offset));
+            addKeyboardProps(span);
+
+            tokenSpans.push(span);
+            tokenSpans.push(delimiters[i]?.includes('\n')
+                ? document.createElement('br')
+                : document.createTextNode(delimiters[i] || ' '),
+            );
+
+            cumulativeOffset += word.length + (delimiters[i]?.length || 0);
+        });
     }
 
     messageLogprobs.forEach((tokenData, i) => {
@@ -101,7 +135,7 @@ function renderAlternativeTokensView() {
         span.text(text);
         span.addClass('logprobs_output_token');
         span.addClass('logprobs_tint_' + (i % TINTS));
-        span.click(() => onSelectedTokenChanged(tokenData, span));
+        span.on('click', () => onSelectedTokenChanged(tokenData, span));
         addKeyboardProps(span);
         tokenSpans.push(...withVirtualWhitespace(token, span));
     });
@@ -129,6 +163,10 @@ function addKeyboardProps(element) {
 /**
  * renderTopLogprobs renders the top logprobs subview with the currently
  * selected token highlighted. If no token is selected, the subview is hidden.
+ *
+ * Callers:
+ * - renderAlternativeTokensView, to render the entire view
+ * - onSelectedTokenChanged, to update the view when a token is selected
  */
 function renderTopLogprobs() {
     $('#logprobs_top_logprobs_hint').hide();
@@ -150,8 +188,7 @@ function renderTopLogprobs() {
                 const probability = Math.exp(log);
                 sum += probability;
                 return [text, probability, log];
-            }
+            } else {
-            else {
                 return [text, log, null];
             }
         });
@@ -167,15 +204,15 @@ function renderTopLogprobs() {
             container.addClass('selected');
         }
 
-        const tokenText = $('<span></span>').text(`${toVisibleWhitespace(token)}`);
+        const tokenText = $('<span></span>').text(`${toVisibleWhitespace(token.toString())}`);
-        const percentText = $('<span></span>').text(`${(probability * 100).toFixed(2)}%`);
+        const percentText = $('<span></span>').text(`${(+probability * 100).toFixed(2)}%`);
         container.append(tokenText, percentText);
         if (log) {
             container.attr('title', `logarithm: ${log}`);
         }
         addKeyboardProps(container);
         if (token !== '<others>') {
-            container.click(() => onAlternativeClicked(state.selectedTokenLogprobs, token));
+            container.on('click', () => onAlternativeClicked(state.selectedTokenLogprobs, token.toString()));
         } else {
             container.prop('disabled', true);
         }
@@ -192,11 +229,10 @@ function renderTopLogprobs() {
 }
 
 /**
- * onSelectedTokenChanged is called when the user clicks on a token in the
+ * User clicks on a token in the token output view. It updates the selected token state
- * token output view. It updates the selected token state and re-renders the
+ * and re-renders the top logprobs view, or deselects the token if it was already selected.
- * top logprobs view, or deselects the token if it was already selected.
  * @param {TokenLogprobs} logprobs - logprob data for the selected token
- * @param {Element} span - target span node that was clicked
+ * @param {Node|JQuery} span - target span node that was clicked
  */
 function onSelectedTokenChanged(logprobs, span) {
     $('.logprobs_output_token.selected').removeClass('selected');
@@ -223,7 +259,10 @@ function onAlternativeClicked(tokenLogprobs, alternative) {
     }
 
     if (getGeneratingApi() === 'openai') {
-        return callPopup('<h3>Feature unavailable</h3><p>Due to API limitations, rerolling a token is not supported with OpenAI. Try switching to a different API.</p>', 'text');
+        const title = t`Feature unavailable`;
+        const message = t`Due to API limitations, rerolling a token is not supported with OpenAI. Try switching to a different API.`;
+        const content = `<h3>${title}</h3><p>${message}</p>`;
+        return callGenericPopup(content, POPUP_TYPE.TEXT);
     }
 
     const { messageLogprobs, continueFrom } = getActiveMessageLogprobData();
@@ -234,79 +273,29 @@ function onAlternativeClicked(tokenLogprobs, alternative) {
 
     const prefix = continueFrom || '';
     const prompt = prefix + tokens.join('');
-    const messageId = chat.length - 1;
+    addGeneration(prompt);
-    createSwipe(messageId, prompt);
-
-    $('.swipe_right:last').click(); // :see_no_evil:
-
-    Generate('continue').then(_ => void _);
 }
 
 /**
- * getTextBeforeClickedWord retrieves the portion of text within a span
+ * User clicks on the reroll button in the token output view, or on a word in the
- * that appears before the word clicked by the user. Using the x and y
+ * prefix. Retrieve the prefix for the current message and truncate it at the
- * coordinates from a PointerEvent, this function identifies the exact
+ * offset for the selected word. Then request a `continue` completion from the
- * word clicked and returns the text preceding it within the span.
+ * model with the new prompt.
  *
- * If the clicked position does not resolve to a valid word or text node,
+ * If no offset is provided, the entire prefix will be rerolled.
- * the entire span text is returned as a fallback.
  *
- * @param {PointerEvent} event - The click event containing the x and y coordinates.
+ * @param {number} offset - index of the token in the prefix to reroll from
- * @param {string} spanText - The full text content of the span element.
+ * @returns {void}
- * @returns {string} The text before the clicked word, or the entire span text as fallback.
+ * @param offset
  */
-function getTextBeforeClickedWord(event, spanText) {
+function onPrefixClicked(offset = undefined) {
-    const x = event.clientX;
-    const y = event.clientY;
-    const range = document.caretRangeFromPoint(x, y);
-
-    if (range && range.startContainer.nodeType === Node.TEXT_NODE) {
-        const textNode = range.startContainer;
-        const offset = range.startOffset;
-
-        // Get the full text content of the text node
-        const text = textNode.nodeValue;
-
-        // Find the boundaries of the clicked word
-        const start = text.lastIndexOf(' ', offset - 1) + 1;
-
-        // Return the text before the clicked word
-        return text.slice(0, start);
-    }
-
-    // If we can't determine the exact word, return the full span text as a fallback
-    return spanText;
-}
-
-
-/**
- * onPrefixClicked is called when the user clicks on the carried-over prefix
- * in the token output view. It allows them to reroll the last 'continue'
- * completion with none of the output generated from it, in case they don't
- * like the results.
- *
- * If the user holds the Ctrl key while clicking, only the portion of text
- * before the clicked word is retained as the prefix for rerolling
- */
-function onPrefixClicked() {
     if (!checkGenerateReady()) {
         return;
     }
 
-    const { continueFrom } = getActiveMessageLogprobData();
+    const { continueFrom } = getActiveMessageLogprobData() || {};
-    const messageId = chat.length - 1;
+    const prefix = continueFrom ? continueFrom.substring(0, offset) : '';
-
+    addGeneration(prefix);
-    // Check if Ctrl key is pressed during the click
-    let prefix = continueFrom || '';
-    if (event.ctrlKey) {
-        // Ctrl is pressed - use the text before the clicked word
-        prefix = getTextBeforeClickedWord(event, continueFrom);
-    }
-
-    // Use the determined `prefix`
-    createSwipe(messageId, prefix);
-    $('.swipe_right:last').click();
-    Generate('continue').then(_ => void _);
 }
 
 function checkGenerateReady() {
@@ -317,6 +306,22 @@ function checkGenerateReady() {
     return true;
 }
 
+/**
+ * Generates a new swipe as a continuation of the given prompt, when user selects
+ * an alternative token or rerolls from a prefix.
+ *
+ * @param prompt
+ */
+function addGeneration(prompt) {
+    const messageId = chat.length - 1;
+    if (prompt && prompt.length > 0) {
+        createSwipe(messageId, prompt);
+        $('.swipe_right:last').trigger('click');
+        void Generate('continue');
+    } else {
+        $('.swipe_right:last').trigger('click');
+    }
+}
 
 /**
  * onToggleLogprobsPanel is called when the user performs an action that toggles
@@ -356,15 +361,14 @@ function onToggleLogprobsPanel() {
 }
 
 /**
- * createSwipe appends a new swipe to the target chat message with the given
+ * Appends a new swipe to the target chat message with the given text.
- * text.
  * @param {number} messageId - target chat message ID
  * @param {string} prompt - initial prompt text which will be continued
  */
 function createSwipe(messageId, prompt) {
     // need to call `cleanUpMessage` on our new prompt, because we were working
     // with raw model output and our new prompt is missing trimming/macro replacements
-    const cleanedPrompt = cleanUpMessage(prompt, false, false);
+    const cleanedPrompt = cleanUpMessage(prompt, false, false, true);
 
     const msg = chat[messageId];
     const newSwipeInfo = {
@@ -399,10 +403,11 @@ function toVisibleWhitespace(input) {
  * after the span node if its token begins or ends with whitespace in order to
  * allow text to wrap despite whitespace characters being replaced with a dot.
  * @param {string} text - token text being evaluated for whitespace
- * @param {Element} span - target span node to be wrapped
+ * @param {Node|JQuery} span - target span node to be wrapped
- * @returns {Element[]} array of nodes to be appended to the DOM
+ * @returns {NodeArray} - array of nodes to be appended to the parent element
  */
 function withVirtualWhitespace(text, span) {
+    /** @type {NodeArray} */
     const result = [span];
     if (text.match(/^\s/)) {
         result.unshift(document.createTextNode('\u200b'));
@@ -430,12 +435,16 @@ function withVirtualWhitespace(text, span) {
 }
 
 /**
- * saveLogprobsForActiveMessage receives an array of TokenLogprobs objects
+ * Receives the top logprobs for each token in a message and associates it with the active message.
- * representing the top logprobs for each token in a message and associates it
+ *
- * with the active message.
+ * Ensure the active message has been updated and rendered before calling this function
+ * or the logprobs data will be saved to the wrong message.
+ *
+ * Callers:
+ * - Generate:onSuccess via saveLogprobsForActiveMessage, for non-streaming text completion
+ * - StreamingProcessor:onFinishStreaming, for streaming text completion
+ * - sendOpenAIRequest, for non-streaming chat completion
  *
- * **Ensure the active message has been updated and rendered before calling
- * this function or the logprobs data will be saved to the wrong message.**
  * @param {TokenLogprobs[]} logprobs - array of logprobs data for each token
  * @param {string | null} continueFrom  - for 'continue' generations, the prompt
  */
@@ -445,7 +454,10 @@ export function saveLogprobsForActiveMessage(logprobs, continueFrom) {
         return;
     }
 
-    convertTokenIdLogprobsToText(logprobs);
+    // NovelAI only returns token IDs in logprobs data; convert to text tokens in-place
+    if (getGeneratingApi() === 'novel') {
+        convertTokenIdLogprobsToText(logprobs);
+    }
 
     const msgId = chat.length - 1;
     /** @type {MessageLogprobData} */
@@ -491,17 +503,18 @@ function getActiveMessageLogprobData() {
     return state.messageLogprobs.get(hash) || null;
 }
 
+
 /**
- * convertLogprobTokenIdsToText mutates the given logprobs data's topLogprobs
+ * convertLogprobTokenIdsToText replaces token IDs in logprobs data with text tokens,
- * field keyed by token text instead of token ID. This is only necessary for
+ * for APIs that return token IDs instead of text tokens, to wit: NovelAI.
- * APIs which only return token IDs in their logprobs data; for others this
+ *
- * function is a no-op.
  * @param {TokenLogprobs[]} input - logprobs data with numeric token IDs
  */
 function convertTokenIdLogprobsToText(input) {
     const api = getGeneratingApi();
     if (api !== 'novel') {
-        return input;
+        // should have been checked by the caller
+        throw new Error('convertTokenIdLogprobsToText should only be called for NovelAI');
     }
 
     const tokenizerId = getTokenizerBestMatch(api);
@@ -512,7 +525,8 @@ function convertTokenIdLogprobsToText(input) {
     )));
 
     // Submit token IDs to tokenizer to get token text, then build ID->text map
-    const { chunks } = decodeTextTokens(tokenizerId, tokenIds);
+    // noinspection JSCheckFunctionSignatures - mutates input in-place
+    const { chunks } = decodeTextTokens(tokenizerId, tokenIds.map(parseInt));
     const tokenIdText = new Map(tokenIds.map((id, i) => [id, chunks[i]]));
 
     // Fixup logprobs data with token text
@@ -525,9 +539,10 @@ function convertTokenIdLogprobsToText(input) {
 }
 
 export function initLogprobs() {
+    REROLL_BUTTON.hide();
     const debouncedRender = debounce(renderAlternativeTokensView);
-    $('#logprobsViewerClose').click(onToggleLogprobsPanel);
+    $('#logprobsViewerClose').on('click', onToggleLogprobsPanel);
-    $('#option_toggle_logprobs').click(onToggleLogprobsPanel);
+    $('#option_toggle_logprobs').on('click', onToggleLogprobsPanel);
     eventSource.on(event_types.CHAT_CHANGED, debouncedRender);
     eventSource.on(event_types.CHARACTER_MESSAGE_RENDERED, debouncedRender);
     eventSource.on(event_types.IMPERSONATE_READY, debouncedRender);

public/scripts/openai.js

@@ -4165,7 +4165,7 @@ async function onModelChange() {
             $('#openai_max_context').attr('max', unlocked_max);
         } else if (oai_settings.mistralai_model.includes('codestral-mamba')) {
             $('#openai_max_context').attr('max', max_256k);
-        } else if (['mistral-large-2407', 'mistral-large-latest'].includes(oai_settings.mistralai_model)) {
+        } else if (['mistral-large-2407', 'mistral-large-2411', 'mistral-large-latest'].includes(oai_settings.mistralai_model)) {
             $('#openai_max_context').attr('max', max_128k);
         } else if (oai_settings.mistralai_model.includes('mistral-nemo')) {
             $('#openai_max_context').attr('max', max_128k);
@@ -4764,6 +4764,8 @@ export function isImageInliningSupported() {
         'pixtral-12b-latest',
         'pixtral-12b',
         'pixtral-12b-2409',
+        'pixtral-large-latest',
+        'pixtral-large-2411',
     ];
 
     switch (oai_settings.chat_completion_source) {

public/scripts/slash-commands.js

@@ -2083,7 +2083,10 @@ async function buttonsCallback(args, text) {
             let popup;
 
             const buttonContainer = document.createElement('div');
-            buttonContainer.classList.add('flex-container', 'flexFlowColumn', 'wide100p', 'm-t-1');
+            buttonContainer.classList.add('flex-container', 'flexFlowColumn', 'wide100p');
+
+            const scrollableContainer = document.createElement('div');
+            scrollableContainer.classList.add('scrollable-buttons-container');
 
             for (const [result, button] of resultToButtonMap) {
                 const buttonElement = document.createElement('div');
@@ -2096,9 +2099,16 @@ async function buttonsCallback(args, text) {
                 buttonContainer.appendChild(buttonElement);
             }
 
+            scrollableContainer.appendChild(buttonContainer);
+
             const popupContainer = document.createElement('div');
             popupContainer.innerHTML = safeValue;
-            popupContainer.appendChild(buttonContainer);
+            popupContainer.appendChild(scrollableContainer);
+
+            // Ensure the popup uses flex layout
+            popupContainer.style.display = 'flex';
+            popupContainer.style.flexDirection = 'column';
+            popupContainer.style.maxHeight = '80vh'; // Limit the overall height of the popup
 
             popup = new Popup(popupContainer, POPUP_TYPE.TEXT, '', { okButton: 'Cancel', allowVerticalScrolling: true });
             popup.show()

public/scripts/textgen-models.js

@@ -23,29 +23,42 @@ export let openRouterModels = [];
 const OPENROUTER_PROVIDERS = [
     'OpenAI',
     'Anthropic',
-    'HuggingFace',
     'Google',
-    'Mancer',
+    'Google AI Studio',
-    'Mancer 2',
+    'Groq',
+    'SambaNova',
+    'Cohere',
+    'Mistral',
     'Together',
+    'Together 2',
+    'Fireworks',
     'DeepInfra',
+    'Lepton',
+    'Novita',
+    'Avian',
+    'Lambda',
     'Azure',
     'Modal',
     'AnyScale',
     'Replicate',
     'Perplexity',
     'Recursal',
-    'Fireworks',
-    'Mistral',
-    'Groq',
-    'Cohere',
-    'Lepton',
     'OctoAI',
-    'Novita',
-    'Lynn',
-    'Lynn 2',
     'DeepSeek',
     'Infermatic',
+    'AI21',
+    'Featherless',
+    'Inflection',
+    'xAI',
+    '01.AI',
+    'HuggingFace',
+    'Mancer',
+    'Mancer 2',
+    'Hyperbolic',
+    'Hyperbolic 2',
+    'Lynn 2',
+    'Lynn',
+    'Reflection',
 ];
 
 export async function loadOllamaModels(data) {

public/style.css

@@ -9,6 +9,7 @@
 @import url(css/logprobs.css);
 @import url(css/accounts.css);
 @import url(css/tags.css);
+@import url(css/scrollable-button.css);
 
 :root {
     --doc-height: 100%;

src/endpoints/stable-diffusion.js

@@ -7,7 +7,7 @@ import sanitize from 'sanitize-filename';
 import { sync as writeFileAtomicSync } from 'write-file-atomic';
 import FormData from 'form-data';
 
-import { getBasicAuthHeader, delay } from '../util.js';
+import { delay, getBasicAuthHeader, tryParse } from '../util.js';
 import { jsonParser } from '../express-common.js';
 import { readSecret, SECRET_KEYS } from './secrets.js';
 
@@ -19,7 +19,7 @@ import { readSecret, SECRET_KEYS } from './secrets.js';
 function getComfyWorkflows(directories) {
     return fs
         .readdirSync(directories.comfyWorkflows)
-        .filter(file => file[0] != '.' && file.toLowerCase().endsWith('.json'))
+        .filter(file => file[0] !== '.' && file.toLowerCase().endsWith('.json'))
         .sort(Intl.Collator().compare);
 }
 
@@ -67,8 +67,7 @@ router.post('/upscalers', jsonParser, async (request, response) => {
 
             /** @type {any} */
             const data = await result.json();
-            const names = data.map(x => x.name);
+            return data.map(x => x.name);
-            return names;
         }
 
         async function getLatentUpscalers() {
@@ -88,8 +87,7 @@ router.post('/upscalers', jsonParser, async (request, response) => {
 
             /** @type {any} */
             const data = await result.json();
-            const names = data.map(x => x.name);
+            return data.map(x => x.name);
-            return names;
         }
 
         const [upscalers, latentUpscalers] = await Promise.all([getUpscalerModels(), getLatentUpscalers()]);
@@ -241,8 +239,7 @@ router.post('/set-model', jsonParser, async (request, response) => {
                     'Authorization': getBasicAuthHeader(request.body.auth),
                 },
             });
-            const data = await result.json();
+            return await result.json();
-            return data;
         }
 
         const url = new URL(request.body.url);
@@ -274,7 +271,7 @@ router.post('/set-model', jsonParser, async (request, response) => {
 
             const progress = progressState['progress'];
             const jobCount = progressState['state']['job_count'];
-            if (progress == 0.0 && jobCount === 0) {
+            if (progress === 0.0 && jobCount === 0) {
                 break;
             }
 
@@ -412,8 +409,19 @@ comfy.post('/models', jsonParser, async (request, response) => {
         }
         /** @type {any} */
         const data = await result.json();
-        return response.send(data.CheckpointLoaderSimple.input.required.ckpt_name[0].map(it => ({ value: it, text: it })));
+
-    } catch (error) {
+        const ckpts = data.CheckpointLoaderSimple.input.required.ckpt_name[0].map(it => ({ value: it, text: it })) || [];
+        const unets = data.UNETLoader.input.required.unet_name[0].map(it => ({ value: it, text: `UNet: ${it}` })) || [];
+
+        // load list of GGUF unets from diffusion_models if the loader node is available
+        const ggufs = data.UnetLoaderGGUF?.input.required.unet_name[0].map(it => ({ value: it, text: `GGUF: ${it}` })) || [];
+        const models = [...ckpts, ...unets, ...ggufs];
+
+        // make the display names of the models somewhat presentable
+        models.forEach(it => it.text = it.text.replace(/\.[^.]*$/, '').replace(/_/g, ' '));
+
+        return response.send(models);
+    } catch (error)     {
         console.log(error);
         return response.sendStatus(500);
     }
@@ -527,7 +535,8 @@ comfy.post('/generate', jsonParser, async (request, response) => {
             body: request.body.prompt,
         });
         if (!promptResult.ok) {
-            throw new Error('ComfyUI returned an error.');
+            const text = await promptResult.text();
+            throw new Error('ComfyUI returned an error.', { cause: tryParse(text) });
         }
 
         /** @type {any} */
@@ -550,7 +559,13 @@ comfy.post('/generate', jsonParser, async (request, response) => {
             await delay(100);
         }
         if (item.status.status_str === 'error') {
-            throw new Error('ComfyUI generation did not succeed.');
+            // Report node tracebacks if available
+            const errorMessages = item.status?.messages
+                ?.filter(it => it[0] === 'execution_error')
+                .map(it => it[1])
+                .map(it => `${it.node_type} [${it.node_id}] ${it.exception_type}: ${it.exception_message}`)
+                .join('\n') || '';
+            throw new Error(`ComfyUI generation did not succeed.\n\n${errorMessages}`.trim());
         }
         const imgInfo = Object.keys(item.outputs).map(it => item.outputs[it].images).flat()[0];
         const imgUrl = new URL(request.body.url);
@@ -560,11 +575,12 @@ comfy.post('/generate', jsonParser, async (request, response) => {
         if (!imgResponse.ok) {
             throw new Error('ComfyUI returned an error.');
         }
-        const imgBuffer = await imgResponse.buffer();
+        const imgBuffer = await imgResponse.arrayBuffer();
-        return response.send(imgBuffer.toString('base64'));
+        return response.send(Buffer.from(imgBuffer).toString('base64'));
     } catch (error) {
-        console.log(error);
+        console.log('ComfyUI error:', error);
-        return response.sendStatus(500);
+        response.status(500).send(error.message);
+        return response;
     }
 });