fixing claude max token testing

litellm/llms/anthropic.py

@@ -60,6 +60,8 @@ def completion(
             prompt += f"{AnthropicConstants.HUMAN_PROMPT.value}{message['content']}"
     prompt += f"{AnthropicConstants.AI_PROMPT.value}"
     max_tokens_to_sample = optional_params.get("max_tokens_to_sample", 256) # required anthropic param, default to 256 if user does not provide an input
+    if max_tokens_to_sample != 256: # not default - print for testing 
+        print_verbose(f"LiteLLM.Anthropic: Max Tokens Set")
     data = {
         "model": model,
         "prompt": prompt,

litellm/tests/test_completion.py

@@ -3,7 +3,7 @@ import traceback
 from dotenv import load_dotenv
 
 load_dotenv()
-import os
+import os, io
 
 sys.path.insert(
     0, os.path.abspath("../..")
@@ -51,6 +51,11 @@ def test_completion_claude():
 
 def test_completion_claude_max_tokens():
     try:
+        litellm.set_verbose = True
+        # Redirect stdout
+        old_stdout = sys.stdout
+        sys.stdout = new_stdout = io.StringIO()
+
         # test setting max tokens for claude-2
         user_message = "tell me everything about YC - be verbose"
         messages = [{"content": user_message, "role": "user"}]
@@ -58,15 +63,17 @@ def test_completion_claude_max_tokens():
         response = completion(
             model="claude-instant-1", messages=messages, max_tokens=1200
         )
+        # Restore stdout
+        sys.stdout = old_stdout
+        output = new_stdout.getvalue().strip()
         # Add any assertions here to check the response
-        print(response)
-        text_response = response['choices'][0]['message']['content']
-        print(len(text_response))
-        assert(len(text_response) > 2000)
-        print(response.response_ms)
+        if "LiteLLM.Anthropic: Max Tokens Set" not in output:
+            raise Exception("Required log message not found!")
+        print(f"response: {response}")
+        litellm.set_verbose = False
     except Exception as e:
         pytest.fail(f"Error occurred: {e}")
-# test_completion_claude_max_tokens()
+test_completion_claude_max_tokens()
 
 # def test_completion_oobabooga():
 #     try:

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "litellm"
-version = "0.1.792"
+version = "0.1.793"
 description = "Library to easily interface with LLM API providers"
 authors = ["BerriAI"]
 license = "MIT License"