calculate time stats based on backend url

llm_server/database.py

@@ -132,39 +132,39 @@ def average_column_for_model(table_name, column_name, model_name):
     return result[0]
 
 
-def weighted_average_column_for_model(table_name, column_name, model_name, backend_name, exclude_zeros: bool = False):
+def weighted_average_column_for_model(table_name, column_name, model_name, backend_name, backend_url, exclude_zeros: bool = False):
     conn = sqlite3.connect(opts.database_path)
     cursor = conn.cursor()
-    cursor.execute(f"SELECT DISTINCT model, backend_mode FROM {table_name}")
-    models_backends = [(row[0], row[1]) for row in cursor.fetchall()]
+    # cursor.execute(f"SELECT DISTINCT model, backend_mode FROM {table_name}")
+    # models_backends = [(row[0], row[1]) for row in cursor.fetchall()]
+    #
+    # model_averages = {}
+    # for model, backend in models_backends:
+    #     if backend != backend_name:
+    #         continue
+    cursor.execute(f"SELECT {column_name}, ROWID FROM {table_name} WHERE model = ? AND backend_mode = ? AND backend_url = ? ORDER BY ROWID DESC", (model_name, backend_name, backend_url))
+    results = cursor.fetchall()
 
-    model_averages = {}
-    for model, backend in models_backends:
-        if backend != backend_name:
+    # if not results:
+    #     continue
+
+    total_weight = 0
+    weighted_sum = 0
+    for i, (value, rowid) in enumerate(results):
+        if value is None or (exclude_zeros and value == 0):
             continue
-        cursor.execute(f"SELECT {column_name}, ROWID FROM {table_name} WHERE model = ? AND backend_mode = ? ORDER BY ROWID DESC", (model, backend))
-        results = cursor.fetchall()
+        weight = i + 1
+        total_weight += weight
+        weighted_sum += weight * value
 
-        if not results:
-            continue
+    # if total_weight == 0:
+    #     continue
 
-        total_weight = 0
-        weighted_sum = 0
-        for i, (value, rowid) in enumerate(results):
-            if value is None or (exclude_zeros and value == 0):
-                continue
-            weight = i + 1
-            total_weight += weight
-            weighted_sum += weight * value
-
-        if total_weight == 0:
-            continue
-
-        model_averages[(model, backend)] = weighted_sum / total_weight
+    calculated_avg = weighted_sum / total_weight
 
     conn.close()
 
-    return model_averages.get((model_name, backend_name))
+    return calculated_avg
 
 
 def sum_column(table_name, column_name):

llm_server/threads.py

@@ -1,9 +1,6 @@
 import time
-from datetime import datetime
 from threading import Thread
 
-import requests
-
 from llm_server import opts
 from llm_server.database import weighted_average_column_for_model
 from llm_server.llm.info import get_running_model
@@ -56,13 +53,13 @@ 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, exclude_zeros=True) or 0
+            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
             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, exclude_zeros=True) or 0
+            average_output_tokens = weighted_average_column_for_model('prompts', 'response_tokens', opts.running_model, opts.mode, opts.backend_url, exclude_zeros=True) or 0
             redis.set('average_output_tokens', average_output_tokens)
 
             # overall = average_column_for_model('prompts', 'response_tokens', opts.running_model)