fix proompters_1_min, other minor changes

llm_server/routes/stats.py

@@ -18,19 +18,17 @@ class SemaphoreCheckerThread(Thread):
         Thread.__init__(self)
         self.semaphore = semaphore
         self.values = collections.deque(maxlen=60)
+        self.prev_semaphore_value = self.semaphore._value
         self.daemon = True
 
     def run(self):
         global proompters_1_min
         while True:
-            # If the semaphore value is less than the maximum, a prompter has sent a prompt
-            if opts.concurrent_gens > self.semaphore._value:
+            current_semaphore_value = self.semaphore._value
+            if current_semaphore_value < opts.concurrent_gens and current_semaphore_value != self.prev_semaphore_value:
                 self.values.append(1)
             else:
                 self.values.append(0)
+            self.prev_semaphore_value = current_semaphore_value
             proompters_1_min = sum(self.values)
             time.sleep(1)
-
-
-thread = SemaphoreCheckerThread(concurrent_semaphore)
-thread.start()

llm_server/routes/v1/generate.py

@@ -64,15 +64,14 @@ def generate():
             backend_response = safe_list_get(response_json_body.get('results', []), 0, {}).get('text')
             if not backend_response:
                 if opts.mode == 'oobabooga':
-                    backend_response = format_sillytavern_err(f'Backend ({opts.mode}) returned an empty string. This can happen when your parameters are incorrect. Make sure your context size is no greater than {opts.context_size}.', 'error')
+                    backend_response = format_sillytavern_err(
+                        f'Backend (oobabooga) returned an empty string. This can happen when your parameters are incorrect. Make sure your context size is no greater than {opts.context_size}. Furthermore, oobabooga does not support concurrent requests so all users have to wait in line and the backend server may have glitched for a moment. Please try again.',
+                        'error')
                     response_json_body['results'][0]['text'] = backend_response
                 else:
                     raise Exception
 
             log_prompt(opts.database_path, client_ip, token, request_json_body['prompt'], backend_response, parameters, dict(request.headers), response.status_code)
-
-            print(response_json_body)
-
             return jsonify({
                 **response_json_body
             }), 200

llm_server/routes/v1/proxy.py

@@ -14,8 +14,8 @@ from ...llm.info import get_running_model
 
 
 @bp.route('/stats', methods=['GET'])
-@cache.cached(timeout=60, query_string=True)
-@cache_control(60)
+@cache.cached(timeout=5, query_string=True)
+@cache_control(5)
 def get_stats():
     model_list = get_running_model()  # will return False when the fetch fails
     if isinstance(model_list, bool):

server.py

@@ -2,16 +2,15 @@ import os
 import sys
 from pathlib import Path
 
-import tiktoken
-from flask import Flask, current_app, jsonify
+from flask import Flask, jsonify
 
 from llm_server import opts
 from llm_server.config import ConfigLoader
 from llm_server.database import init_db
 from llm_server.helpers import resolve_path
-from llm_server.llm.info import get_running_model
 from llm_server.routes.cache import cache
 from llm_server.routes.helpers.http import cache_control
+from llm_server.routes.stats import SemaphoreCheckerThread, concurrent_semaphore
 from llm_server.routes.v1 import bp
 
 script_path = os.path.dirname(os.path.realpath(__file__))
@@ -48,6 +47,8 @@ opts.concurrent_gens = config['concurrent_gens']
 opts.frontend_api_client = config['frontend_api_client']
 opts.context_size = config['token_limit']
 
+SemaphoreCheckerThread(concurrent_semaphore).start()
+
 app = Flask(__name__)
 cache.init_app(app)
 cache.clear()  # clear redis cache
@@ -71,4 +72,4 @@ def fallback(first=None, rest=None):
 
 
 if __name__ == "__main__":
-    app.run(host='0.0.0.0', debug=True)
+    app.run(host='0.0.0.0')