Merge pull request #6987 from oobabooga/dev

Merge dev branch

css/chat_style-Dark.css

@@ -2,8 +2,8 @@
     display: grid;
     align-items: start;
     grid-template-columns: 60px minmax(0, 1fr);
-    padding-bottom: 14px;
+    padding-bottom: 22px;
-    padding-top: 14px;
+    padding-top: 6px;
     font-size: 18px;
     font-family: Roboto, Arial, sans-serif; /* Modern font */
     line-height: 1.5;

css/chat_style-TheEncrypted777.css

@@ -4,8 +4,8 @@
     display: grid;
     align-items: start;
     grid-template-columns: 60px minmax(0, 1fr);
-    padding-bottom: 14px;
+    padding-bottom: 21px;
-    padding-top: 14px;
+    padding-top: 7px;
     font-size: 18px;
     font-family: 'Noto Sans', Arial, sans-serif;
     line-height: 1.428571429;

css/chat_style-cai-chat-square.css

@@ -16,7 +16,7 @@
 }
 
 .message {
-    padding-bottom: 1em;
+    padding-bottom: 1.5em;
-    padding-top: 1em;
+    padding-top: 0.5em;
     grid-template-columns: 70px minmax(0, 1fr);
 }

css/chat_style-cai-chat.css

@@ -2,8 +2,8 @@
     display: grid;
     align-items: start;
     grid-template-columns: 60px minmax(0, 1fr);
-    padding-bottom: 1em;
+    padding-bottom: 1.5em;
-    padding-top: 1em;
+    padding-top: 0.5em;
     font-size: 15px;
     font-family: 'Noto Sans', Helvetica, Arial, sans-serif;
     line-height: 22.5px !important;

css/chat_style-messenger.css

@@ -1,6 +1,6 @@
 .message {
-    padding-bottom: 12.5px;
+    padding-bottom: 22px;
-    padding-top: 12.5px;
+    padding-top: 3px;
     font-size: 15px;
     font-family: 'Noto Sans', Helvetica, Arial, sans-serif;
     line-height: 1.428571429;

css/chat_style-wpp.css

@@ -1,6 +1,6 @@
 .message {
-    padding-bottom: 12.5px;
+    padding-bottom: 22px;
-    padding-top: 12.5px;
+    padding-top: 3px;
     font-size: 15px;
     font-family: 'Noto Sans', Helvetica, Arial, sans-serif;
     line-height: 1.428571429;

css/main.css

@@ -390,7 +390,7 @@ div.svelte-362y77>*, div.svelte-362y77>.form>* {
     margin-left: auto;
     margin-right: auto;
     flex: 1;
-    overflow-y: auto;
+    overflow-y: hidden;
     display: flex;
     flex-direction: column;
     word-break: break-word;

js/main.js

@@ -152,7 +152,7 @@ const observer = new MutationObserver(function(mutations) {
   }
 
   const chatElement = document.getElementById("chat");
-  if (chatElement) {
+  if (chatElement && chatElement.getAttribute("data-mode") === "instruct") {
     const messagesContainer = chatElement.querySelector(".messages");
     const lastChild = messagesContainer?.lastElementChild;
     const prevSibling = lastChild?.previousElementSibling;

modules/html_generator.py

@@ -347,7 +347,7 @@ remove_button = f'<button class="footer-button footer-remove-button" title="Remo
 
 
 def generate_instruct_html(history):
-    output = f'<style>{instruct_css}</style><div class="chat" id="chat"><div class="messages">'
+    output = f'<style>{instruct_css}</style><div class="chat" id="chat" data-mode="instruct"><div class="messages">'
 
     for i in range(len(history['visible'])):
         row_visible = history['visible'][i]

modules/models_settings.py

@@ -72,6 +72,7 @@ def get_model_metadata(model):
                 model_settings['compress_pos_emb'] = metadata[k]
             elif k.endswith('block_count'):
                 model_settings['gpu_layers'] = metadata[k] + 1
+                model_settings['max_gpu_layers'] = metadata[k] + 1
 
         if 'tokenizer.chat_template' in metadata:
             template = metadata['tokenizer.chat_template']
@@ -153,6 +154,9 @@ def get_model_metadata(model):
     for pat in settings:
         if re.match(pat.lower(), Path(model).name.lower()):
             for k in settings[pat]:
+                if k == 'n_gpu_layers':
+                    k = 'gpu_layers'
+
                 model_settings[k] = settings[pat][k]
 
     # Load instruction template if defined by name rather than by value
@@ -450,17 +454,31 @@ def update_gpu_layers_and_vram(loader, model, gpu_layers, ctx_size, cache_type,
     max_layers = gpu_layers
 
     if auto_adjust:
-        # Get max layers from model metadata
+        # Get model settings including user preferences
         model_settings = get_model_metadata(model)
-        max_layers = model_settings.get('gpu_layers', gpu_layers)
 
-        # Auto-adjust based on available VRAM
+        # Get the true maximum layers
-        available_vram = get_nvidia_free_vram()
+        max_layers = model_settings.get('max_gpu_layers', model_settings.get('gpu_layers', gpu_layers))
-        if available_vram > 0:
+
-            tolerance = 906
+        # Check if this is a user-saved setting
-            current_layers = max_layers
+        user_config = shared.user_config
-            while current_layers > 0 and estimate_vram(model, current_layers, ctx_size, cache_type) > available_vram - tolerance:
+        model_regex = Path(model).name + '$'
-                current_layers -= 1
+        has_user_setting = model_regex in user_config and 'gpu_layers' in user_config[model_regex]
+
+        if has_user_setting:
+            # For user settings, just use the current value (which already has user pref)
+            # but ensure the slider maximum is correct
+            current_layers = gpu_layers  # Already has user setting
+        else:
+            # No user setting, auto-adjust from the maximum
+            current_layers = max_layers  # Start from max
+
+            # Auto-adjust based on available VRAM
+            available_vram = get_nvidia_free_vram()
+            if available_vram > 0:
+                tolerance = 906
+                while current_layers > 0 and estimate_vram(model, current_layers, ctx_size, cache_type) > available_vram - tolerance:
+                    current_layers -= 1
 
     # Calculate VRAM with current layers
     vram_usage = estimate_vram(model, current_layers, ctx_size, cache_type)

modules/ui_model_menu.py

@@ -318,7 +318,7 @@ def get_initial_vram_info():
 def get_initial_gpu_layers_max():
     if shared.model_name != 'None' and shared.args.loader == 'llama.cpp':
         model_settings = get_model_metadata(shared.model_name)
-        return model_settings.get('gpu_layers', 256)
+        return model_settings.get('max_gpu_layers', model_settings.get('gpu_layers', 256))
 
     return 256
 

modules/utils.py

@@ -137,7 +137,7 @@ def get_available_models():
 
     model_dirs = sorted(model_dirs, key=natural_keys)
 
-    return ['None'] + filtered_gguf_files + model_dirs
+    return filtered_gguf_files + model_dirs
 
 
 def get_available_ggufs():