Merge pull request #70 from LlamaEnjoyer/qol_tweaks

Add auto session naming and context tracking qol features.

backend/sessions.py

@@ -188,6 +188,31 @@ class Session:
         self.save()
 
 
+    def _create_session_name_from_text(self, text, max_length=30):
+        """Creates a session name from the first message text.
+        
+        Args:
+            text: The message text to create name from
+            max_length: Maximum length of generated name
+            
+        Returns:
+            A name string, truncated with ellipsis if needed
+        """
+        if not text:
+            return "Unnamed session"
+        
+        processed_text = text.split(": ", 1)[-1].strip()  # Cache split result
+        words = processed_text.split()
+        name = ""
+        for word in words:
+            if len(name) + len(word) + 4 <= max_length:  # +4 for "..." and space
+                name += word + " "
+            else:
+                name = name.strip() + "..."
+                break
+        
+        return name.strip() or "Unnamed session"
+
     def user_input(self, data):
         prompt_format = prompt_formats[self.settings["prompt_format"]]()
         input_text = data["user_input_text"]
@@ -197,6 +222,13 @@ class Session:
         if prompt_format.is_instruct(): prefix = ""
         else: prefix = self.settings["roles"][0] + ": "
         new_block["text"] = prefix + input_text
+        
+        # Auto-rename session if this is the first message
+        if len(self.history) == 0:
+            self.name = self._create_session_name_from_text(input_text)
+            if session_list is not None and self.session_uuid in session_list:
+                session_list[self.session_uuid] = (self.name, session_list[self.session_uuid][1])
+        
         self.history.append(new_block)
         self.save()
         return new_block
@@ -659,6 +691,8 @@ class Session:
         meta["gen_speed"] = generated_tokens / (mt.stages["gen"] + 1e-8)
         meta["overflow"] = max_new_tokens if generated_tokens == max_new_tokens else 0
         meta["canceled"] = abort_event.is_set()
+        meta["context_tokens"] = context_ids.shape[-1] + save_tokens.shape[-1]  # Total tokens in context
+        meta["max_seq_len"] = model.config.max_seq_len  # Maximum sequence length
         new_block["meta"] = meta
 
         # Save response block
@@ -720,4 +754,3 @@ class Session:
                 self.history[i] = block
                 break
         self.save()
-

static/chat.js

@@ -688,9 +688,12 @@ class ChatBlock {
             let ptps = this.block.meta.prompt_speed.toFixed(2)
             if (this.block.meta.prompt_speed > 50000) ptps = "∞";
 
-            let html = "prompt: " + this.block.meta.prompt_tokens.toFixed(0) + " tokens, " + ptps + " tokens/s";
+            let contextPercent = (this.block.meta.context_tokens / this.block.meta.max_seq_len * 100).toFixed(0);
+            let html = "prompt: " + this.block.meta.prompt_tokens.toFixed(0) + " tokens, " + ptps + " tokens/s ";
             html += " ⁄ ";
-            html += "response: " + this.block.meta.gen_tokens.toFixed(0) + " tokens, " + this.block.meta.gen_speed.toFixed(2) + " tokens/s";
+            html += "response: " + this.block.meta.gen_tokens.toFixed(0) + " tokens, " + this.block.meta.gen_speed.toFixed(2) + " tokens/s ";
+            html += " ⁄ ";
+            html += "context: " + contextPercent + "% full";
             p.innerHTML = html;
             this.textBlock.appendChild(p);
         }
@@ -855,4 +858,4 @@ class ChatBlock {
         block.parent.currentStreamingBlock = block;
         block.parent.getModelResponse(block.block.block_uuid, block.block.text);
     }
-}
+}

static/models.js

@@ -442,6 +442,8 @@ export class ModelView {
     }
 
     loadModel() {
+        const controller = new AbortController();
+        const signal = controller.signal;
 
         overlay.loadingOverlay.setProgress(0, 1);
         overlay.pageOverlay.setMode("loading");
@@ -456,10 +458,18 @@ export class ModelView {
             }, 10000)
         });
 
+        overlay.loadingOverlay.onCancel = () => {
+            controller.abort();
+            overlay.pageOverlay.setMode();
+            this.error_message = "Loading cancelled";
+            this.updateView();
+        };
+
         let fetchRequest = fetch("/api/load_model", {
             method: "POST",
             headers: { "Content-Type": "application/json", },
-            body: JSON.stringify(packet)
+            body: JSON.stringify(packet),
+            signal: signal
         });
 
         const self = this;

static/overlay.css

@@ -44,4 +44,8 @@
     border-radius: 0px;
     text-align: center;
     line-height: 30px;
-}
+}
+
+.overlay .textbutton {
+    margin-top: 25px;
+}

static/overlay.js

@@ -1,4 +1,6 @@
 import * as util from "./util.js";
+import * as controls from "./controls.js";
+import * as globals from "./globals.js";
 
 class PageOverlay {
     constructor() {
@@ -60,6 +62,25 @@ class LoadingOverlay extends Overlay {
 
         this.bar = util.newDiv(null, "progressbar-bar");
         this.box.appendChild(this.bar);
+
+        this.cancelButton = new controls.Button("✖ Cancel", () => {
+            // First call the cancel handler to abort the fetch request
+            if (this.onCancel) this.onCancel();
+            
+            // Then unload the model to release GPU memory
+            fetch("/api/unload_model")
+            .then(response => response.json())
+            .then(json => {
+                if (json.result == "ok") {
+                    globals.g.loadedModelUUID = null;
+                    globals.g.failedModelUUID = null;
+                }
+            });
+        });
+        this.cancelButton.setEnabled(true);
+        this.element.appendChild(this.cancelButton.element);
+        
+        this.onCancel = null;
     }
 
     setProgress(a, b) {