reorganize stats, add 24 hr proompters, adjust logging when error

llm_server/database.py

@@ -110,3 +110,14 @@ def sum_column(table_name, column_name):
     result = cursor.fetchone()
     conn.close()
     return result[0] if result[0] else 0
+
+
+def get_distinct_ips_24h():
+    # Get the current time and subtract 24 hours (in seconds)
+    past_24_hours = int(time.time()) - 24 * 60 * 60
+    conn = sqlite3.connect(opts.database_path)
+    cur = conn.cursor()
+    cur.execute("SELECT COUNT(DISTINCT ip) FROM prompts WHERE timestamp >= ?", (past_24_hours,))
+    result = cur.fetchone()
+    conn.close()
+    return result[0] if result else 0

llm_server/routes/v1/generate.py

@@ -75,7 +75,7 @@ def generate():
         else:
             raise Exception
 
-        log_prompt(client_ip, token, request_json_body['prompt'], backend_response, elapsed_time, parameters, dict(request.headers), response if response else 0)
+        log_prompt(client_ip, token, request_json_body['prompt'], backend_response, None, parameters, dict(request.headers), response if response else 0)
         return jsonify({
             'code': 500,
             'error': 'failed to reach backend',
@@ -90,13 +90,13 @@ def generate():
             if opts.mode == 'oobabooga':
                 backend_err = True
                 backend_response = format_sillytavern_err(
-                    f'Backend (oobabooga) returned an empty string. This is usually due to an error on the backend during inference. Make sure your context size is no greater than {opts.context_size}. Please try again.',
+                    f'Backend (oobabooga) returned an empty string. This is usually due to an error on the backend during inference. Please check your parameters and try again.',
                     'error')
                 response_json_body['results'][0]['text'] = backend_response
             else:
                 raise Exception
 
-        log_prompt(client_ip, token, request_json_body['prompt'], backend_response if not backend_err else '', elapsed_time, parameters, dict(request.headers), response.status_code)
+        log_prompt(client_ip, token, request_json_body['prompt'], backend_response if not backend_err else '', elapsed_time if not backend_err else None, parameters, dict(request.headers), response.status_code)
         return jsonify({
             **response_json_body
         }), 200

llm_server/routes/v1/generate_stats.py

@@ -2,7 +2,7 @@ import time
 from datetime import datetime
 
 from llm_server import opts
-from llm_server.database import sum_column
+from llm_server.database import get_distinct_ips_24h, sum_column
 from llm_server.helpers import deep_sort
 from llm_server.llm.info import get_running_model
 from llm_server.netdata import get_power_states
@@ -28,7 +28,7 @@ def generate_stats():
     #     waits = [elapsed for end, elapsed in t]
     #     estimated_wait = int(sum(waits) / len(waits))
 
-    proompters_in_queue = len(priority_queue) + get_active_gen_workers()
+    proompters_in_queue = len(priority_queue)
     average_tps = float(redis.get('average_tps'))
 
     if opts.average_generation_time_mode == 'database':
@@ -48,9 +48,15 @@ def generate_stats():
 
     output = {
         'stats': {
-            'proompts_in_queue': proompters_in_queue,
-            'proompters_1_min': SemaphoreCheckerThread.proompters_1_min,
-            'proompts': get_total_proompts() if opts.show_num_prompts else None,
+            'proompters': {
+                '1_min': SemaphoreCheckerThread.proompters_1_min,
+                '24_hrs': get_distinct_ips_24h(),
+            },
+            'proompts': {
+                'processing': get_active_gen_workers(),
+                'queued': proompters_in_queue,
+                'total': get_total_proompts() if opts.show_num_prompts else None,
+            },
             'uptime': int((datetime.now() - server_start_time).total_seconds()) if opts.show_uptime else None,
             'average_generation_elapsed_sec': average_generation_time,
             'average_tps': average_tps,