Merge pull request #206 from Webifi/main

Sticky tool drawer, try to track down streaming responses getting stuck

src/app.scss

@@ -537,6 +537,12 @@ aside.menu.main-menu .menu-expanse {
   min-width: 60px;
   min-height: 1.3em;
 }
+.button-pack {
+  display: block;
+  position: sticky;
+  top: calc(var(--sectionPaddingTop) + 10px);
+  padding-bottom: 0.1px;
+}
 .button-pack .button {
   display: block;
   margin: 4px;
@@ -597,23 +603,27 @@ aside.menu.main-menu .menu-expanse {
   width: var(--chatToolDrawerSize);
   visibility: visible;
   max-height: 300%;
+  overflow: unset;
 }
 .user-message:hover .tool-drawer,
 .user-message.editing .tool-drawer {
   width: var(--chatToolDrawerSize);
   visibility: visible;
   max-height: 300%;
+  overflow: unset;
 }
 
 .assistant-message:hover .tool-drawer-mask,
 .assistant-message.editing .tool-drawer {
   width: var(--chatToolDrawerSize);
   visibility: visible;
+  overflow: unset;
 }
 .user-message:hover .tool-drawer-mask,
 .user-message.editing .tool-drawer {
   width: var(--chatToolDrawerSize);
   visibility: visible;
+  overflow: unset;
 }
 .assistant-message:hover, .assistant-message.editing {
   border-top-right-radius: 0px !important;

src/lib/ChatCompletionResponse.svelte

@@ -104,9 +104,9 @@ export class ChatCompletionResponse {
           completion_tokens: 0,
           total_tokens: 0
         } as Usage
-        message.usage.completion_tokens += response.usage.completion_tokens
+        message.usage.completion_tokens += response?.usage?.completion_tokens || 0
-        message.usage.prompt_tokens = response.usage.prompt_tokens + (this.offsetTotals?.prompt_tokens || 0)
+        message.usage.prompt_tokens = (response?.usage?.prompt_tokens || 0) + (this.offsetTotals?.prompt_tokens || 0)
-        message.usage.total_tokens = response.usage.total_tokens + (this.offsetTotals?.total_tokens || 0)
+        message.usage.total_tokens = (response?.usage?.total_tokens || 0) + (this.offsetTotals?.total_tokens || 0)
       } else {
         message.content = choice.message.content
         message.usage = response.usage
@@ -124,7 +124,7 @@ export class ChatCompletionResponse {
   updateFromAsyncResponse (response: Response) {
     let completionTokenCount = 0
     this.setModel(response.model)
-    if (!response.choices) {
+    if (!response.choices || response?.error) {
       return this.updateFromError(response?.error?.message || 'unexpected streaming response from API')
     }
     response.choices?.forEach((choice, i) => {
@@ -174,7 +174,11 @@ export class ChatCompletionResponse {
     setTimeout(() => this.finish(), 250) // give others a chance to signal the finish first
   }
 
-  updateFromClose (): void {
+  updateFromClose (force: boolean = false): void {
+    if (!this.finished && !this.error && !this.messages?.find(m => m.content)) {
+      if (!force) return setTimeout(() => this.updateFromClose(true), 250) as any
+      return this.updateFromError('Unexpected connection termination')
+    }
     setTimeout(() => this.finish(), 250) // give others a chance to signal the finish first
   }
 

src/lib/ChatRequest.svelte

@@ -202,6 +202,8 @@ export class ChatRequest {
           _this.controller = new AbortController()
           const signal = _this.controller.signal
           const abortListener = (e:Event) => {
+            _this.updating = false
+            _this.updatingMessage = ''
             chatResponse.updateFromError('User aborted request.')
             signal.removeEventListener('abort', abortListener)
           }
@@ -245,6 +247,8 @@ export class ChatRequest {
                 }
               },
               onclose () {
+                _this.updating = false
+                _this.updatingMessage = ''
                 chatResponse.updateFromClose()
               },
               onerror (err) {
@@ -260,6 +264,8 @@ export class ChatRequest {
                 }
               }
             }).catch(err => {
+              _this.updating = false
+              _this.updatingMessage = ''
               chatResponse.updateFromError(err.message)
             })
           } else {

src/lib/Storage.svelte

@@ -200,9 +200,9 @@
       }
       chat.usage[model] = total
     }
-    total.completion_tokens += usage.completion_tokens
+    total.completion_tokens += usage?.completion_tokens || 0
-    total.prompt_tokens += usage.prompt_tokens
+    total.prompt_tokens += usage?.prompt_tokens || 0
-    total.total_tokens += usage.total_tokens
+    total.total_tokens += usage?.total_tokens || 0
     chatsStorage.set(chats)
   }
 
@@ -218,9 +218,9 @@
       }
       chat.usage[model] = total
     }
-    total.completion_tokens -= usage.completion_tokens
+    total.completion_tokens -= usage?.completion_tokens || 0
-    total.prompt_tokens -= usage.prompt_tokens
+    total.prompt_tokens -= usage?.prompt_tokens || 0
-    total.total_tokens -= usage.total_tokens
+    total.total_tokens -= usage?.total_tokens || 0
     chatsStorage.set(chats)
   }