Merge branch 'oobabooga:dev' into dev

README.md

@@ -13,7 +13,7 @@
 
 # Text Generation Web UI
 
-A Gradio web UI for Large Language Models.
+Run AI chatbots like ChatGPT on your own computer. **100% private and offline** – no subscriptions, no API fees, zero telemetry. Just download, unzip, and run.
 
 [Try the Deep Reason extension](https://oobabooga.gumroad.com/l/deep_reason)
 
@@ -21,38 +21,35 @@ A Gradio web UI for Large Language Models.
 |:---:|:---:|
 |![Image1](https://github.com/oobabooga/screenshots/raw/main/DEFAULT-3.5.png) | ![Image2](https://github.com/oobabooga/screenshots/raw/main/PARAMETERS-3.5.png) |
 
-## 🔥 News
-
-- The project now supports **image generation**! Including Z-Image-Turbo, 4bit/8bit quantization, `torch.compile`, and LLM-generated prompt variations ([tutorial](https://github.com/oobabooga/text-generation-webui/wiki/Image-Generation-Tutorial)).
-
 ## Features
 
-- Supports multiple local text generation backends, including [llama.cpp](https://github.com/ggerganov/llama.cpp), [Transformers](https://github.com/huggingface/transformers), [ExLlamaV3](https://github.com/turboderp-org/exllamav3), [ExLlamaV2](https://github.com/turboderp-org/exllamav2), and [TensorRT-LLM](https://github.com/NVIDIA/TensorRT-LLM) (the latter via its own [Dockerfile](https://github.com/oobabooga/text-generation-webui/blob/main/docker/TensorRT-LLM/Dockerfile)).
-- Easy setup: Choose between **portable builds** (zero setup, just unzip and run) for GGUF models on Windows/Linux/macOS, or the one-click installer that creates a self-contained `installer_files` directory.
 - 100% offline and private, with zero telemetry, external resources, or remote update requests.
 - **File attachments**: Upload text files, PDF documents, and .docx documents to talk about their contents.
 - **Vision (multimodal models)**: Attach images to messages for visual understanding ([tutorial](https://github.com/oobabooga/text-generation-webui/wiki/Multimodal-Tutorial)).
 - **Image generation**: A dedicated tab for `diffusers` models like **Z-Image-Turbo**. Features 4-bit/8-bit quantization and a persistent gallery with metadata ([tutorial](https://github.com/oobabooga/text-generation-webui/wiki/Image-Generation-Tutorial)).
 - **Web search**: Optionally search the internet with LLM-generated queries to add context to the conversation.
-- Aesthetic UI with dark and light themes.
-- Syntax highlighting for code blocks and LaTeX rendering for mathematical expressions.
+- Aesthetic UI with dark/light themes, syntax highlighting, and LaTeX rendering.
+- Edit messages, navigate between message versions, and branch conversations at any point.
+- Switch between models without restarting, with automatic GPU layer allocation.
+- Free-form text generation in the Notebook tab without being limited to chat turns.
 - `instruct` mode for instruction-following (like ChatGPT), and `chat-instruct`/`chat` modes for talking to custom characters.
 - Automatic prompt formatting using Jinja2 templates. You don't need to ever worry about prompt formats.
-- Edit messages, navigate between message versions, and branch conversations at any point.
 - Multiple sampling parameters and generation options for sophisticated text generation control.
-- Switch between different models in the UI without restarting.
-- Automatic GPU layers for GGUF models (on NVIDIA GPUs).
-- Free-form text generation in the Notebook tab without being limited to chat turns.
+- Supports multiple backends including [llama.cpp](https://github.com/ggerganov/llama.cpp), [Transformers](https://github.com/huggingface/transformers), [ExLlamaV3](https://github.com/turboderp-org/exllamav3), [ExLlamaV2](https://github.com/turboderp-org/exllamav2), and [TensorRT-LLM](https://github.com/NVIDIA/TensorRT-LLM).
 - OpenAI-compatible API with Chat and Completions endpoints, including tool-calling support – see [examples](https://github.com/oobabooga/text-generation-webui/wiki/12-%E2%80%90-OpenAI-API#examples).
 - Extension support, with numerous built-in and user-contributed extensions available. See the [wiki](https://github.com/oobabooga/text-generation-webui/wiki/07-%E2%80%90-Extensions) and [extensions directory](https://github.com/oobabooga/text-generation-webui-extensions) for details.
 
+## 🔥 News
+
+- The project now supports **image generation**! Including Z-Image-Turbo, 4bit/8bit quantization, `torch.compile`, and LLM-generated prompt variations ([tutorial](https://github.com/oobabooga/text-generation-webui/wiki/Image-Generation-Tutorial)).
+
 ## How to install
 
 #### ✅ Option 1: Portable builds (get started in 1 minute)
 
 No installation needed – just download, unzip and run. All dependencies included.
 
-Compatible with GGUF (llama.cpp) models on Windows, Linux, and macOS.
+Compatible with GGUF (llama.cpp) models on Windows, Linux, and macOS. [Check what models fit your hardware](https://huggingface.co/spaces/oobabooga/accurate-gguf-vram-calculator).
 
 Download from here: **https://github.com/oobabooga/text-generation-webui/releases**
 

css/main.css

@@ -1645,7 +1645,7 @@ button:focus {
 }
 
 #user-description textarea {
-    height: calc(100vh - 231px) !important;
+    height: calc(100vh - 334px) !important;
     min-height: 90px !important;
 }
 

js/highlightjs/highlightjs-copy.min.js

@@ -1 +1,84 @@
-class CopyButtonPlugin{constructor(options={}){self.hook=options.hook;self.callback=options.callback;self.lang=options.lang||document.documentElement.lang||"en"}"after:highlightElement"({el,text}){let button=Object.assign(document.createElement("button"),{innerHTML:locales[lang]?.[0]||"Copy",className:"hljs-copy-button"});button.dataset.copied=false;el.parentElement.classList.add("hljs-copy-wrapper");el.parentElement.appendChild(button);el.parentElement.style.setProperty("--hljs-theme-background",window.getComputedStyle(el).backgroundColor);button.onclick=function(){if(!navigator.clipboard)return;let newText=text;if(hook&&typeof hook==="function"){newText=hook(text,el)||text}navigator.clipboard.writeText(newText).then(function(){button.innerHTML=locales[lang]?.[1]||"Copied!";button.dataset.copied=true;let alert=Object.assign(document.createElement("div"),{role:"status",className:"hljs-copy-alert",innerHTML:locales[lang]?.[2]||"Copied to clipboard"});el.parentElement.appendChild(alert);setTimeout(()=>{button.innerHTML=locales[lang]?.[0]||"Copy";button.dataset.copied=false;el.parentElement.removeChild(alert);alert=null},2e3)}).then(function(){if(typeof callback==="function")return callback(newText,el)})}}}if(typeof module!="undefined"){module.exports=CopyButtonPlugin}const locales={en:["Copy","Copied!","Copied to clipboard"],es:["Copiar","¡Copiado!","Copiado al portapapeles"],fr:["Copier","Copié !","Copié dans le presse-papier"],de:["Kopieren","Kopiert!","In die Zwischenablage kopiert"],ja:["コピー","コピーしました！","クリップボードにコピーしました"],ko:["복사","복사됨!","클립보드에 복사됨"],ru:["Копировать","Скопировано!","Скопировано в буфер обмена"],zh:["复制","已复制!","已复制到剪贴板"],"zh-tw":["複製","已複製!","已複製到剪貼簿"]};
+function fallbackCopyToClipboard(text) {
+  return new Promise((resolve, reject) => {
+    const textArea = document.createElement("textarea");
+    textArea.value = text;
+    textArea.style.position = "fixed";
+    textArea.style.left = "-9999px";
+    textArea.style.top = "-9999px";
+    document.body.appendChild(textArea);
+    textArea.focus();
+    textArea.select();
+    try {
+      const successful = document.execCommand("copy");
+      document.body.removeChild(textArea);
+      successful ? resolve() : reject();
+    } catch (err) {
+      document.body.removeChild(textArea);
+      reject(err);
+    }
+  });
+}
+
+class CopyButtonPlugin {
+  constructor(options = {}) {
+    self.hook = options.hook;
+    self.callback = options.callback;
+    self.lang = options.lang || document.documentElement.lang || "en";
+  }
+  "after:highlightElement"({ el, text }) {
+    let button = Object.assign(document.createElement("button"), {
+      innerHTML: locales[lang]?.[0] || "Copy",
+      className: "hljs-copy-button",
+    });
+    button.dataset.copied = false;
+    el.parentElement.classList.add("hljs-copy-wrapper");
+    el.parentElement.appendChild(button);
+    el.parentElement.style.setProperty(
+      "--hljs-theme-background",
+      window.getComputedStyle(el).backgroundColor,
+    );
+    button.onclick = function () {
+      let newText = text;
+      if (hook && typeof hook === "function") {
+        newText = hook(text, el) || text;
+      }
+      const copyPromise =
+        navigator.clipboard && window.isSecureContext
+          ? navigator.clipboard.writeText(newText)
+          : fallbackCopyToClipboard(newText);
+      copyPromise.then(function () {
+          button.innerHTML = locales[lang]?.[1] || "Copied!";
+          button.dataset.copied = true;
+          let alert = Object.assign(document.createElement("div"), {
+            role: "status",
+            className: "hljs-copy-alert",
+            innerHTML: locales[lang]?.[2] || "Copied to clipboard",
+          });
+          el.parentElement.appendChild(alert);
+          setTimeout(() => {
+            button.innerHTML = locales[lang]?.[0] || "Copy";
+            button.dataset.copied = false;
+            el.parentElement.removeChild(alert);
+            alert = null;
+          }, 2e3);
+        })
+        .then(function () {
+          if (typeof callback === "function") return callback(newText, el);
+        });
+    };
+  }
+}
+if (typeof module != "undefined") {
+  module.exports = CopyButtonPlugin;
+}
+const locales = {
+  en: ["Copy", "Copied!", "Copied to clipboard"],
+  es: ["Copiar", "¡Copiado!", "Copiado al portapapeles"],
+  fr: ["Copier", "Copié !", "Copié dans le presse-papier"],
+  de: ["Kopieren", "Kopiert!", "In die Zwischenablage kopiert"],
+  ja: ["コピー", "コピーしました！", "クリップボードにコピーしました"],
+  ko: ["복사", "복사됨!", "클립보드에 복사됨"],
+  ru: ["Копировать", "Скопировано!", "Скопировано в буфер обмена"],
+  zh: ["复制", "已复制!", "已复制到剪贴板"],
+  "zh-tw": ["複製", "已複製!", "已複製到剪貼簿"],
+};

modules/chat.py

@@ -32,7 +32,12 @@ from modules.text_generation import (
     get_encoded_length,
     get_max_prompt_length
 )
-from modules.utils import delete_file, get_available_characters, save_file
+from modules.utils import (
+    delete_file,
+    get_available_characters,
+    get_available_users,
+    save_file
+)
 from modules.web_search import add_web_search_attachments
 
 
@@ -1647,6 +1652,150 @@ def delete_character(name, instruct=False):
         delete_file(Path(f'user_data/characters/{name}.{extension}'))
 
 
+def generate_user_pfp_cache(user):
+    """Generate cached profile picture for user"""
+    cache_folder = Path(shared.args.disk_cache_dir)
+    if not cache_folder.exists():
+        cache_folder.mkdir()
+
+    for path in [Path(f"user_data/users/{user}.{extension}") for extension in ['png', 'jpg', 'jpeg']]:
+        if path.exists():
+            original_img = Image.open(path)
+            # Define file paths
+            pfp_path = Path(f'{cache_folder}/pfp_me.png')
+
+            # Save thumbnail
+            thumb = make_thumbnail(original_img)
+            thumb.save(pfp_path, format='PNG')
+            logger.info(f'User profile picture cached to "{pfp_path}"')
+
+            return str(pfp_path)
+
+    return None
+
+
+def load_user(user_name, name1, user_bio):
+    """Load user profile from YAML file"""
+    picture = None
+
+    filepath = None
+    for extension in ["yml", "yaml", "json"]:
+        filepath = Path(f'user_data/users/{user_name}.{extension}')
+        if filepath.exists():
+            break
+
+    if filepath is None or not filepath.exists():
+        logger.error(f"Could not find the user \"{user_name}\" inside user_data/users. No user has been loaded.")
+        raise ValueError
+
+    with open(filepath, 'r', encoding='utf-8') as f:
+        file_contents = f.read()
+
+    extension = filepath.suffix[1:]  # Remove the leading dot
+    data = json.loads(file_contents) if extension == "json" else yaml.safe_load(file_contents)
+
+    # Clear existing user picture cache
+    cache_folder = Path(shared.args.disk_cache_dir)
+    pfp_path = Path(f"{cache_folder}/pfp_me.png")
+    if pfp_path.exists():
+        pfp_path.unlink()
+
+    # Generate new picture cache
+    picture = generate_user_pfp_cache(user_name)
+
+    # Get user name
+    if 'name' in data and data['name'] != '':
+        name1 = data['name']
+
+    # Get user bio
+    if 'user_bio' in data:
+        user_bio = data['user_bio']
+
+    return name1, user_bio, picture
+
+
+def generate_user_yaml(name, user_bio):
+    """Generate YAML content for user profile"""
+    data = {
+        'name': name,
+        'user_bio': user_bio,
+    }
+
+    return yaml.dump(data, sort_keys=False, width=float("inf"))
+
+
+def save_user(name, user_bio, picture, filename):
+    """Save user profile to YAML file"""
+    if filename == "":
+        logger.error("The filename is empty, so the user will not be saved.")
+        return
+
+    # Ensure the users directory exists
+    users_dir = Path('user_data/users')
+    users_dir.mkdir(parents=True, exist_ok=True)
+
+    data = generate_user_yaml(name, user_bio)
+    filepath = Path(f'user_data/users/{filename}.yaml')
+    save_file(filepath, data)
+
+    path_to_img = Path(f'user_data/users/{filename}.png')
+    if picture is not None:
+        # Copy the image file from its source path to the users folder
+        shutil.copy(picture, path_to_img)
+        logger.info(f'Saved user profile picture to {path_to_img}.')
+
+
+def delete_user(name):
+    """Delete user profile files"""
+    # Check for user data files
+    for extension in ["yml", "yaml", "json"]:
+        delete_file(Path(f'user_data/users/{name}.{extension}'))
+
+    # Check for user image files
+    for extension in ["png", "jpg", "jpeg"]:
+        delete_file(Path(f'user_data/users/{name}.{extension}'))
+
+
+def update_user_menu_after_deletion(idx):
+    """Update user menu after a user is deleted"""
+    users = get_available_users()
+    if len(users) == 0:
+        # Create a default user if none exist
+        save_user('You', '', None, 'Default')
+        users = get_available_users()
+
+    idx = min(int(idx), len(users) - 1)
+    idx = max(0, idx)
+    return gr.update(choices=users, value=users[idx])
+
+
+def handle_user_menu_change(state):
+    """Handle user menu selection change"""
+    try:
+        name1, user_bio, picture = load_user(state['user_menu'], state['name1'], state['user_bio'])
+
+        return [
+            name1,
+            user_bio,
+            picture
+        ]
+    except Exception as e:
+        logger.error(f"Failed to load user '{state['user_menu']}': {e}")
+        return [
+            state['name1'],
+            state['user_bio'],
+            None
+        ]
+
+
+def handle_save_user_click(name1):
+    """Handle save user button click"""
+    return [
+        name1,
+        gr.update(visible=True)
+    ]
+
+
 def jinja_template_from_old_format(params, verbose=False):
     MASTER_TEMPLATE = """
 {%- set ns = namespace(found=false) -%}

modules/html_generator.py

@@ -108,91 +108,64 @@ def replace_blockquote(m):
     return m.group().replace('\n', '\n> ').replace('\\begin{blockquote}', '').replace('\\end{blockquote}', '')
 
 
+# Thinking block format definitions: (start_tag, end_tag, content_start_tag)
+# Use None for start_tag to match from beginning (end-only formats should be listed last)
+THINKING_FORMATS = [
+    ('<think>', '</think>', None),
+    ('<|channel|>analysis<|message|>', '<|end|>', '<|start|>assistant<|channel|>final<|message|>'),
+    ('<seed:think>', '</seed:think>', None),
+    ('<|think|>', '<|end|>', '<|content|>'),  # Solar Open
+    (None, '</think>', None),  # End-only variant (e.g., Qwen3-next)
+]
+
+
 def extract_thinking_block(string):
     """Extract thinking blocks from the beginning of a string."""
     if not string:
         return None, string
 
-    THINK_START_TAG = "&lt;think&gt;"
-    THINK_END_TAG = "&lt;/think&gt;"
+    for start_tag, end_tag, content_tag in THINKING_FORMATS:
+        end_esc = html.escape(end_tag)
+        content_esc = html.escape(content_tag) if content_tag else None
 
-    # Look for think tag first
-    start_pos = string.find(THINK_START_TAG)
-    end_pos = string.find(THINK_END_TAG)
-
-    # If think tags found, use existing logic
-    if start_pos != -1 or end_pos != -1:
-        # handle missing start or end tags
-        if start_pos == -1:
+        if start_tag is None:
+            # End-only format: require end tag, start from beginning
+            end_pos = string.find(end_esc)
+            if end_pos == -1:
+                continue
             thought_start = 0
         else:
-            thought_start = start_pos + len(THINK_START_TAG)
+            # Normal format: require start tag
+            start_esc = html.escape(start_tag)
+            start_pos = string.find(start_esc)
+            if start_pos == -1:
+                continue
+            thought_start = start_pos + len(start_esc)
+            end_pos = string.find(end_esc, thought_start)
+
         if end_pos == -1:
-            thought_end = len(string)
-            content_start = len(string)
-        else:
-            thought_end = end_pos
-            content_start = end_pos + len(THINK_END_TAG)
-        thinking_content = string[thought_start:thought_end]
-        remaining_content = string[content_start:]
-        return thinking_content, remaining_content
-
-    # If think tags not found, try GPT-OSS alternative format
-    ALT_START = "&lt;|channel|&gt;analysis&lt;|message|&gt;"
-    ALT_END = "&lt;|end|&gt;"
-    ALT_CONTENT_START = "&lt;|start|&gt;assistant&lt;|channel|&gt;final&lt;|message|&gt;"
-
-    alt_start_pos = string.find(ALT_START)
-    alt_end_pos = string.find(ALT_END)
-    alt_content_pos = string.find(ALT_CONTENT_START)
-
-    if alt_start_pos != -1 or alt_end_pos != -1:
-        if alt_start_pos == -1:
-            thought_start = 0
-        else:
-            thought_start = alt_start_pos + len(ALT_START)
-
-        # If no explicit end tag but content start exists, use content start as end
-        if alt_end_pos == -1:
-            if alt_content_pos != -1:
-                thought_end = alt_content_pos
-                content_start = alt_content_pos + len(ALT_CONTENT_START)
+            # End tag missing - check if content tag can serve as fallback
+            if content_esc:
+                content_pos = string.find(content_esc, thought_start)
+                if content_pos != -1:
+                    thought_end = content_pos
+                    content_start = content_pos + len(content_esc)
+                else:
+                    thought_end = len(string)
+                    content_start = len(string)
             else:
                 thought_end = len(string)
                 content_start = len(string)
         else:
-            thought_end = alt_end_pos
-            content_start = alt_content_pos + len(ALT_CONTENT_START) if alt_content_pos != -1 else alt_end_pos + len(ALT_END)
+            thought_end = end_pos
+            if content_esc:
+                content_pos = string.find(content_esc, end_pos)
+                content_start = content_pos + len(content_esc) if content_pos != -1 else end_pos + len(end_esc)
+            else:
+                content_start = end_pos + len(end_esc)
 
-        thinking_content = string[thought_start:thought_end]
-        remaining_content = string[content_start:]
-        return thinking_content, remaining_content
+        return string[thought_start:thought_end], string[content_start:]
 
-    # Try seed:think format
-    SEED_START = "&lt;seed:think&gt;"
-    SEED_END = "&lt;/seed:think&gt;"
-
-    seed_start_pos = string.find(SEED_START)
-    seed_end_pos = string.find(SEED_END)
-
-    if seed_start_pos != -1 or seed_end_pos != -1:
-        if seed_start_pos == -1:
-            thought_start = 0
-        else:
-            thought_start = seed_start_pos + len(SEED_START)
-
-        if seed_end_pos == -1:
-            thought_end = len(string)
-            content_start = len(string)
-        else:
-            thought_end = seed_end_pos
-            content_start = seed_end_pos + len(SEED_END)
-
-        thinking_content = string[thought_start:thought_end]
-        remaining_content = string[content_start:]
-        return thinking_content, remaining_content
-
-    # Return if no format is found
     return None, string
 
 

modules/llama_cpp_server.py

@@ -298,10 +298,24 @@ class LlamaServer:
         if "bos_token" in response:
             self.bos_token = response["bos_token"]
 
-    def _find_available_port(self):
-        """Find an available port by letting the OS assign one."""
+    def _is_port_available(self, port):
+        """Check if a port is available for use."""
         with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
-            s.bind(('', 0))  # Bind to port 0 to get an available port
+            try:
+                s.bind(('', port))
+                return True
+            except OSError:
+                return False
+
+    def _find_available_port(self):
+        """Find an available port, preferring main port + 1."""
+        preferred_port = shared.args.api_port + 1
+        if self._is_port_available(preferred_port):
+            return preferred_port
+
+        # Fall back to OS-assigned random port
+        with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+            s.bind(('', 0))
             return s.getsockname()[1]
 
     def _start_server(self):

modules/shared.py

@@ -298,6 +298,7 @@ settings = {
 
     # Character settings
     'character': 'Assistant',
+    'user': 'Default',
     'name1': 'You',
     'name2': 'AI',
     'user_bio': '',

modules/ui.py

@@ -251,6 +251,7 @@ def list_interface_input_elements():
         'chat_style',
         'chat-instruct_command',
         'character_menu',
+        'user_menu',
         'name2',
         'context',
         'greeting',
@@ -353,6 +354,8 @@ def save_settings(state, preset, extensions_list, show_controls, theme_state, ma
     output['preset'] = preset
     output['prompt-notebook'] = state['prompt_menu-default'] if state['show_two_notebook_columns'] else state['prompt_menu-notebook']
     output['character'] = state['character_menu']
+    if 'user_menu' in state and state['user_menu']:
+        output['user'] = state['user_menu']
     output['seed'] = int(output['seed'])
     output['show_controls'] = show_controls
     output['dark_theme'] = True if theme_state == 'dark' else False
@@ -457,6 +460,7 @@ def setup_auto_save():
         'chat_style',
         'chat-instruct_command',
         'character_menu',
+        'user_menu',
         'name1',
         'name2',
         'context',

modules/ui_chat.py

@@ -137,6 +137,12 @@ def create_character_settings_ui():
                     shared.gradio['greeting'] = gr.Textbox(value=shared.settings['greeting'], lines=5, label='Greeting', elem_classes=['add_scrollbar'], elem_id="character-greeting")
 
                 with gr.Tab("User"):
+                    with gr.Row():
+                        shared.gradio['user_menu'] = gr.Dropdown(value=shared.settings['user'], choices=utils.get_available_users(), label='User', elem_id='user-menu', info='Select a user profile.', elem_classes='slim-dropdown')
+                        ui.create_refresh_button(shared.gradio['user_menu'], lambda: None, lambda: {'choices': utils.get_available_users()}, 'refresh-button', interactive=not mu)
+                        shared.gradio['save_user'] = gr.Button('💾', elem_classes='refresh-button', elem_id="save-user", interactive=not mu)
+                        shared.gradio['delete_user'] = gr.Button('🗑️', elem_classes='refresh-button', interactive=not mu)
+
                     shared.gradio['name1'] = gr.Textbox(value=shared.settings['name1'], lines=1, label='Name')
                     shared.gradio['user_bio'] = gr.Textbox(value=shared.settings['user_bio'], lines=10, label='Description', info='Here you can optionally write a description of yourself.', placeholder='{{user}}\'s personality: ...', elem_classes=['add_scrollbar'], elem_id="user-description")
 
@@ -372,3 +378,11 @@ def create_event_handlers():
         gradio('enable_web_search'),
         gradio('web_search_row')
     )
+
+    # User menu event handlers
+    shared.gradio['user_menu'].change(
+        ui.gather_interface_values, gradio(shared.input_elements), gradio('interface_state')).then(
+        chat.handle_user_menu_change, gradio('interface_state'), gradio('name1', 'user_bio', 'your_picture'), show_progress=False)
+
+    shared.gradio['save_user'].click(chat.handle_save_user_click, gradio('name1'), gradio('save_user_filename', 'user_saver'), show_progress=False)
+    shared.gradio['delete_user'].click(lambda: gr.update(visible=True), None, gradio('user_deleter'), show_progress=False)

modules/ui_file_saving.py

@@ -39,6 +39,19 @@ def create_ui():
             shared.gradio['delete_character_cancel'] = gr.Button('Cancel', elem_classes="small-button")
             shared.gradio['delete_character_confirm'] = gr.Button('Delete', elem_classes="small-button", variant='stop', interactive=not mu)
 
+    # User saver/deleter
+    with gr.Group(visible=False, elem_classes='file-saver') as shared.gradio['user_saver']:
+        shared.gradio['save_user_filename'] = gr.Textbox(lines=1, label='File name', info='The user profile will be saved to your user_data/users folder with this base filename.')
+        with gr.Row():
+            shared.gradio['save_user_cancel'] = gr.Button('Cancel', elem_classes="small-button")
+            shared.gradio['save_user_confirm'] = gr.Button('Save', elem_classes="small-button", variant='primary', interactive=not mu)
+
+    with gr.Group(visible=False, elem_classes='file-saver') as shared.gradio['user_deleter']:
+        gr.Markdown('Confirm the user deletion?')
+        with gr.Row():
+            shared.gradio['delete_user_cancel'] = gr.Button('Cancel', elem_classes="small-button")
+            shared.gradio['delete_user_confirm'] = gr.Button('Delete', elem_classes="small-button", variant='stop', interactive=not mu)
+
     # Preset saver
     with gr.Group(visible=False, elem_classes='file-saver') as shared.gradio['preset_saver']:
         shared.gradio['save_preset_filename'] = gr.Textbox(lines=1, label='File name', info='The preset will be saved to your user_data/presets folder with this base filename.')
@@ -69,6 +82,12 @@ def create_event_handlers():
     shared.gradio['save_character_cancel'].click(lambda: gr.update(visible=False), None, gradio('character_saver'), show_progress=False)
     shared.gradio['delete_character_cancel'].click(lambda: gr.update(visible=False), None, gradio('character_deleter'), show_progress=False)
 
+    # User save/delete event handlers
+    shared.gradio['save_user_confirm'].click(handle_save_user_confirm_click, gradio('name1', 'user_bio', 'your_picture', 'save_user_filename'), gradio('user_menu', 'user_saver'), show_progress=False)
+    shared.gradio['delete_user_confirm'].click(handle_delete_user_confirm_click, gradio('user_menu'), gradio('user_menu', 'user_deleter'), show_progress=False)
+    shared.gradio['save_user_cancel'].click(lambda: gr.update(visible=False), None, gradio('user_saver'), show_progress=False)
+    shared.gradio['delete_user_cancel'].click(lambda: gr.update(visible=False), None, gradio('user_deleter'), show_progress=False)
+
 
 def handle_save_preset_confirm_click(filename, contents):
     try:
@@ -165,3 +184,33 @@ def handle_delete_grammar_click(grammar_file):
         "user_data/grammars/",
         gr.update(visible=True)
     ]
+
+
+def handle_save_user_confirm_click(name1, user_bio, your_picture, filename):
+    try:
+        chat.save_user(name1, user_bio, your_picture, filename)
+        available_users = utils.get_available_users()
+        output = gr.update(choices=available_users, value=filename)
+    except Exception:
+        output = gr.update()
+        traceback.print_exc()
+
+    return [
+        output,
+        gr.update(visible=False)
+    ]
+
+
+def handle_delete_user_confirm_click(user):
+    try:
+        index = str(utils.get_available_users().index(user))
+        chat.delete_user(user)
+        output = chat.update_user_menu_after_deletion(index)
+    except Exception:
+        output = gr.update()
+        traceback.print_exc()
+
+    return [
+        output,
+        gr.update(visible=False)
+    ]

modules/utils.py

@@ -219,6 +219,13 @@ def get_available_characters():
     return sorted(set((k.stem for k in paths)), key=natural_keys)
 
 
+def get_available_users():
+    users_dir = Path('user_data/users')
+    users_dir.mkdir(parents=True, exist_ok=True)
+    paths = (x for x in users_dir.iterdir() if x.suffix in ('.json', '.yaml', '.yml'))
+    return sorted(set((k.stem for k in paths)), key=natural_keys)
+
+
 def get_available_instruction_templates():
     path = "user_data/instruction-templates"
     paths = []

requirements/full/requirements.txt

@@ -28,7 +28,7 @@ sentencepiece
 tensorboard
 torchao==0.15.*
 transformers==4.57.*
-triton-windows==3.5.1.post22; platform_system == "Windows"
+triton-windows==3.5.1.post24; platform_system == "Windows"
 tqdm
 wandb
 

requirements/full/requirements_amd.txt

@@ -26,7 +26,7 @@ sentencepiece
 tensorboard
 torchao==0.15.*
 transformers==4.57.*
-triton-windows==3.5.1.post22; platform_system == "Windows"
+triton-windows==3.5.1.post24; platform_system == "Windows"
 tqdm
 wandb
 

requirements/full/requirements_amd_noavx2.txt

@@ -26,7 +26,7 @@ sentencepiece
 tensorboard
 torchao==0.15.*
 transformers==4.57.*
-triton-windows==3.5.1.post22; platform_system == "Windows"
+triton-windows==3.5.1.post24; platform_system == "Windows"
 tqdm
 wandb
 

requirements/full/requirements_apple_intel.txt

@@ -26,7 +26,7 @@ sentencepiece
 tensorboard
 torchao==0.15.*
 transformers==4.57.*
-triton-windows==3.5.1.post22; platform_system == "Windows"
+triton-windows==3.5.1.post24; platform_system == "Windows"
 tqdm
 wandb
 

requirements/full/requirements_apple_silicon.txt

@@ -26,7 +26,7 @@ sentencepiece
 tensorboard
 torchao==0.15.*
 transformers==4.57.*
-triton-windows==3.5.1.post22; platform_system == "Windows"
+triton-windows==3.5.1.post24; platform_system == "Windows"
 tqdm
 wandb
 

requirements/full/requirements_cpu_only.txt

@@ -26,7 +26,7 @@ sentencepiece
 tensorboard
 torchao==0.15.*
 transformers==4.57.*
-triton-windows==3.5.1.post22; platform_system == "Windows"
+triton-windows==3.5.1.post24; platform_system == "Windows"
 tqdm
 wandb
 

requirements/full/requirements_cpu_only_noavx2.txt

@@ -26,7 +26,7 @@ sentencepiece
 tensorboard
 torchao==0.15.*
 transformers==4.57.*
-triton-windows==3.5.1.post22; platform_system == "Windows"
+triton-windows==3.5.1.post24; platform_system == "Windows"
 tqdm
 wandb
 

requirements/full/requirements_noavx2.txt

@@ -28,7 +28,7 @@ sentencepiece
 tensorboard
 torchao==0.15.*
 transformers==4.57.*
-triton-windows==3.5.1.post22; platform_system == "Windows"
+triton-windows==3.5.1.post24; platform_system == "Windows"
 tqdm
 wandb
 

requirements/full/requirements_nowheels.txt

@@ -26,7 +26,7 @@ sentencepiece
 tensorboard
 torchao==0.15.*
 transformers==4.57.*
-triton-windows==3.5.1.post22; platform_system == "Windows"
+triton-windows==3.5.1.post24; platform_system == "Windows"
 tqdm
 wandb
 

user_data/users/Default.yaml

@@ -0,0 +1,2 @@
+name: You
+user_bio: ''