fix(alerting.py): fix datetime comparison logic

litellm/integrations/slack_alerting.py

@@ -12,7 +12,7 @@ from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler
 import datetime
 from pydantic import BaseModel
 from enum import Enum
-from datetime import datetime as dt, timedelta
+from datetime import datetime as dt, timedelta, timezone
 from litellm.integrations.custom_logger import CustomLogger
 import random
 
@@ -32,7 +32,9 @@ class LiteLLMBase(BaseModel):
 
 class SlackAlertingArgs(LiteLLMBase):
     default_daily_report_frequency: int = 12 * 60 * 60  # 12 hours
-    daily_report_frequency: int = int(os.getenv("SLACK_DAILY_REPORT_FREQUENCY", default_daily_report_frequency))
+    daily_report_frequency: int = int(
+        os.getenv("SLACK_DAILY_REPORT_FREQUENCY", default_daily_report_frequency)
+    )
     report_check_interval: int = 5 * 60  # 5 minutes
 
 
@@ -373,8 +375,10 @@ class SlackAlerting(CustomLogger):
             key=lambda i: replaced_failed_values[i],
             reverse=True,
         )[:5]
-        top_5_failed = [index for index in top_5_failed if replaced_failed_values[index] > 0]
-        
+        top_5_failed = [
+            index for index in top_5_failed if replaced_failed_values[index] > 0
+        ]
+
         # find top 5 slowest
         # Replace None values with a placeholder value (-1 in this case)
         placeholder_value = 0
@@ -389,7 +393,9 @@ class SlackAlerting(CustomLogger):
             key=lambda i: replaced_slowest_values[i],
             reverse=True,
         )[:5]
-        top_5_slowest = [index for index in top_5_slowest if replaced_slowest_values[index] > 0]
+        top_5_slowest = [
+            index for index in top_5_slowest if replaced_slowest_values[index] > 0
+        ]
 
         # format alert -> return the litellm model name + api base
         message = f"\n\nHere are today's key metrics 📈: \n\n"
@@ -847,15 +853,19 @@ Model Info:
                 value=_current_time,
             )
         else:
-            # check if current time - interval >= time last sent
-            delta = current_time - timedelta(
-                seconds=self.alerting_args.daily_report_frequency
-            )
-
+            # Check if current time - interval >= time last sent
+            delta_naive = timedelta(seconds=self.alerting_args.daily_report_frequency)
             if isinstance(report_sent, str):
                 report_sent = dt.fromisoformat(report_sent)
 
-            if delta >= report_sent:
+            # Ensure report_sent is an aware datetime object
+            if report_sent.tzinfo is None:
+                report_sent = report_sent.replace(tzinfo=timezone.utc)
+
+            # Calculate delta as an aware datetime object with the same timezone as report_sent
+            delta = report_sent - delta_naive
+
+            if current_time >= delta:
                 # Sneak in the reporting logic here
                 await self.send_daily_reports(router=llm_router)
                 # Also, don't forget to update the report_sent time after sending the report!

litellm/utils.py

@@ -59,9 +59,7 @@ from importlib import resources
 with resources.open_text("litellm.llms.tokenizers", "anthropic_tokenizer.json") as f:
     json_data = json.load(f)
 # Convert to str (if necessary)
-json_str = json.dumps(json_data)
-claude_tokenizer = Tokenizer.from_str(json_str)
-cohere_tokenizer = Tokenizer.from_pretrained("Xenova/c4ai-command-r-v01-tokenizer")
+claude_json_str = json.dumps(json_data)
 import importlib.metadata
 from ._logging import verbose_logger
 from .types.router import LiteLLM_Params
@@ -3856,12 +3854,15 @@ def get_replicate_completion_pricing(completion_response=None, total_time=0.0):
 
 @lru_cache(maxsize=128)
 def _select_tokenizer(model: str):
-    global claude_tokenizer, cohere_tokenizer
     if model in litellm.cohere_models and "command-r" in model:
         # cohere
+        cohere_tokenizer = Tokenizer.from_pretrained(
+            "Xenova/c4ai-command-r-v01-tokenizer"
+        )
         return {"type": "huggingface_tokenizer", "tokenizer": cohere_tokenizer}
     # anthropic
     elif model in litellm.anthropic_models and "claude-3" not in model:
+        claude_tokenizer = Tokenizer.from_str(claude_json_str)
         return {"type": "huggingface_tokenizer", "tokenizer": claude_tokenizer}
     # llama2
     elif "llama-2" in model.lower() or "replicate" in model.lower():

proxy_server_config.yaml

@@ -1,10 +1,16 @@
 model_list:
-  - model_name: gpt-3.5-turbo
+  - model_name: gpt-3.5-turbo-end-user-test
     litellm_params:
       model: gpt-3.5-turbo
       region_name: "eu"
     model_info:
       id: "1"
+  - model_name: gpt-3.5-turbo-end-user-test
+    litellm_params:
+      model: azure/chatgpt-v-2
+      api_base: https://openai-gpt-4-test-v-1.openai.azure.com/
+      api_version: "2023-05-15"
+      api_key: os.environ/AZURE_API_KEY # The `os.environ/` prefix tells litellm to read this from the env. See https://docs.litellm.ai/docs/simple_proxy#load-api-keys-from-vault
   - model_name: gpt-3.5-turbo
     litellm_params:
       model: azure/chatgpt-v-2

tests/test_end_users.py

@@ -153,7 +153,9 @@ async def test_end_user_specific_region():
         )
 
         ## MAKE CALL ##
-        key_gen = await generate_key(session=session, i=0, models=["gpt-3.5-turbo"])
+        key_gen = await generate_key(
+            session=session, i=0, models=["gpt-3.5-turbo-end-user-test"]
+        )
 
         key = key_gen["key"]
 
@@ -162,7 +164,7 @@ async def test_end_user_specific_region():
 
         print("SENDING USER PARAM - {}".format(end_user_obj["user_id"]))
         result = await client.chat.completions.with_raw_response.create(
-            model="gpt-3.5-turbo",
+            model="gpt-3.5-turbo-end-user-test",
             messages=[{"role": "user", "content": "Hey!"}],
             user=end_user_obj["user_id"],
         )