fix two exceptions

llm_server/database/database.py

@@ -1,5 +1,6 @@
 import json
 import time
+import traceback
 
 import llm_server
 from llm_server import opts
@@ -113,9 +114,13 @@ def average_column_for_model(table_name, column_name, model_name):
 def weighted_average_column_for_model(table_name, column_name, model_name, backend_name, backend_url, exclude_zeros: bool = False):
     conn = db_pool.connection()
     cursor = conn.cursor()
+    try:
         try:
             cursor.execute(f"SELECT {column_name}, id FROM {table_name} WHERE model = %s AND backend_mode = %s AND backend_url = %s ORDER BY id DESC", (model_name, backend_name, backend_url,))
             results = cursor.fetchall()
+        except Exception:
+            traceback.print_exc()
+            return -1
 
         total_weight = 0
         weighted_sum = 0

llm_server/routes/v1/generate_stream.py

@@ -70,6 +70,10 @@ def stream(ws):
                     if partial_response.endswith(b'\x00'):
                         json_str = partial_response[:-1].decode()  # Remove the null character and decode the byte string to a string
                         json_obj = json.loads(json_str)
+                        if not len(json_obj['text'][0].split(input_prompt + generated_text)):
+                            # ????
+                            continue
+
                         new = json_obj['text'][0].split(input_prompt + generated_text)[1]
 
                         ws.send(json.dumps({

llm_server/threads.py

@@ -47,12 +47,14 @@ class MainBackgroundThread(Thread):
             # exclude_zeros=True filters out rows where an error message was returned. Previously, if there was an error, 0
             # was entered into the column. The new code enters null instead but we need to be backwards compatible for now.
             average_generation_elapsed_sec = weighted_average_column_for_model('prompts', 'generation_time', opts.running_model, opts.mode, opts.backend_url, exclude_zeros=True) or 0
+            if average_generation_elapsed_sec > -1:
                 redis.set('average_generation_elapsed_sec', average_generation_elapsed_sec)
 
             # overall = average_column_for_model('prompts', 'generation_time', opts.running_model)
             # print(f'Weighted: {average_generation_elapsed_sec}, overall: {overall}')
 
             average_output_tokens = weighted_average_column_for_model('prompts', 'response_tokens', opts.running_model, opts.mode, opts.backend_url, exclude_zeros=True) or 0
+            if average_generation_elapsed_sec > -1:
                 redis.set('average_output_tokens', average_output_tokens)
 
             # overall = average_column_for_model('prompts', 'response_tokens', opts.running_model)