daemon.py

@@ -20,8 +20,8 @@ else:
     config_path = Path(script_path, 'config', 'config.yml')
 
 if __name__ == "__main__":
-    Redis().flushall()
-    print('Flushed Redis.')
+    # Redis().flushall()
+    # print('Flushed Redis.')
 
     success, config, msg = load_config(config_path)
     if not success:

llm_server/workers/inferencer.py

@@ -19,6 +19,8 @@ def worker(backend_url):
         increment_ip_count(client_ip, 'processing_ips')
         incr_active_workers(selected_model, backend_url)
 
+        print('Worker starting processing for', client_ip)
+
         try:
             if not request_json_body:
                 # This was a dummy request from the streaming handlers.

llm_server/workers/threader.py

@@ -44,7 +44,6 @@ def start_background():
     t.start()
     print('Started the console printer.')
 
-    redis_running_models.flush()
     t = Thread(target=cluster_worker)
     t.daemon = True
     t.start()

other/gradio/gradio_chat.py

@@ -12,6 +12,12 @@ if not API_BASE:
     print('Must set the secret variable API_BASE to your https://your-site/api/openai/v1')
     sys.exit(1)
 
+BACKUP_API_BASE = os.getenv('BACKUP_API_BASE')
+if BACKUP_API_BASE:
+    print('Using BACKUP_API_BASE:', BACKUP_API_BASE)
+
+APP_TITLE = os.getenv('APP_TITLE')
+
 # A system prompt can be injected into the very first spot in the context.
 # If the user sends a message that contains the CONTEXT_TRIGGER_PHRASE,
 # the content in CONTEXT_TRIGGER_INJECTION will be injected.
@@ -37,16 +43,22 @@ def stream_response(prompt, history):
     if do_injection or (CONTEXT_TRIGGER_INJECTION and CONTEXT_TRIGGER_PHRASE in prompt):
         messages.insert(0, {'role': 'system', 'content': CONTEXT_TRIGGER_INJECTION})
 
-    try:
-        response = openai.ChatCompletion.create(
-            model='0',
-            messages=messages,
-            temperature=0,
-            max_tokens=300,
-            stream=True
-        )
-    except Exception:
-        raise gr.Error("Failed to reach inference endpoint.")
+    for _ in range(2):
+        try:
+            response = openai.ChatCompletion.create(
+                model='0',
+                messages=messages,
+                temperature=0,
+                max_tokens=300,
+                stream=True
+            )
+            break
+        except Exception:
+            openai.api_base = BACKUP_API_BASE
+            raise gr.Error("Failed to reach inference endpoint.")
+
+    # Go back to the default endpoint
+    openai.api_base = API_BASE
 
     message = ''
     for chunk in response:
@@ -55,8 +67,8 @@ def stream_response(prompt, history):
             yield message
 
 
-examples = ["hello", "hola", "merhaba"]
+examples = ["hello"]
 if CONTEXT_TRIGGER_PHRASE:
     examples.insert(0, CONTEXT_TRIGGER_PHRASE)
 
-gr.ChatInterface(stream_response, examples=examples, title="Chatbot Demo", analytics_enabled=False, cache_examples=False, css='#component-0{height:100%!important}').queue(concurrency_count=3).launch()
+gr.ChatInterface(stream_response, examples=examples, title=APP_TITLE, analytics_enabled=False, cache_examples=False, css='#component-0{height:100%!important}').queue(concurrency_count=1, api_open=False).launch(show_api=False)