Merge pull request #3518 from BerriAI/revert-3478-Issue-#3474-anthropic-roles-alternation-issue

Revert "* feat(factory.py): add support for merging consecutive messages of one role when separated with empty message of another role"

.gitignore

@@ -1,5 +1,4 @@
 .venv
-venv
 .env
 litellm_uuid.txt
 __pycache__/

litellm/llms/prompt_templates/factory.py

@@ -1,23 +1,27 @@
-import json
-import re
-import traceback
-import uuid
-import xml.etree.ElementTree as ET
 from enum import Enum
-from typing import Any, List, Mapping, MutableMapping, Optional, Sequence
+import requests, traceback
-
+import json, re, xml.etree.ElementTree as ET
-import requests
+from jinja2 import Template, exceptions, meta, BaseLoader
-from jinja2 import BaseLoader, Template, exceptions, meta
 from jinja2.sandbox import ImmutableSandboxedEnvironment
-
+from typing import (
+    Any,
+    List,
+    Mapping,
+    MutableMapping,
+    Optional,
+    Sequence,
+)
 import litellm
-from litellm.types.completion import (ChatCompletionFunctionMessageParam,
+from litellm.types.completion import (
-                                      ChatCompletionMessageParam,
+    ChatCompletionUserMessageParam,
-                                      ChatCompletionMessageToolCallParam,
     ChatCompletionSystemMessageParam,
+    ChatCompletionMessageParam,
+    ChatCompletionFunctionMessageParam,
+    ChatCompletionMessageToolCallParam,
     ChatCompletionToolMessageParam,
-                                      ChatCompletionUserMessageParam)
+)
 from litellm.types.llms.anthropic import *
+import uuid
 
 
 def default_pt(messages):
@@ -599,9 +603,8 @@ def construct_tool_use_system_prompt(
 
 
 def convert_url_to_base64(url):
-    import base64
-
     import requests
+    import base64
 
     for _ in range(3):
         try:
@@ -981,7 +984,6 @@ def anthropic_messages_pt(messages: list):
     new_messages = []
     msg_i = 0
     tool_use_param = False
-    merge_with_previous = False
     while msg_i < len(messages):
         user_content = []
         init_msg_i = msg_i
@@ -1014,13 +1016,7 @@ def anthropic_messages_pt(messages: list):
             msg_i += 1
 
         if user_content:
-            if merge_with_previous:
-                new_messages[-1]["content"].extend(user_content)
-                merge_with_previous = False
-            else:
             new_messages.append({"role": "user", "content": user_content})
-        elif msg_i > 0:
-            merge_with_previous = True
 
         assistant_content = []
         ## MERGE CONSECUTIVE ASSISTANT CONTENT ##
@@ -1048,13 +1044,7 @@ def anthropic_messages_pt(messages: list):
             msg_i += 1
 
         if assistant_content:
-            if merge_with_previous:
-                new_messages[-1]["content"].extend(assistant_content)
-                merge_with_previous = False
-            else:
             new_messages.append({"role": "assistant", "content": assistant_content})
-        elif msg_i > 0:
-            merge_with_previous = True
 
         if msg_i == init_msg_i:  # prevent infinite loops
             raise Exception(

litellm/tests/test_completion.py

@@ -1,21 +1,17 @@
-import os
+import sys, os
-import sys
 import traceback
-
 from dotenv import load_dotenv
 
 load_dotenv()
-import io
+import os, io
-import os
 
 sys.path.insert(
     0, os.path.abspath("../..")
 )  # Adds the parent directory to the, system path
 import pytest
-
 import litellm
-from litellm import (RateLimitError, Timeout, completion, completion_cost,
+from litellm import embedding, completion, completion_cost, Timeout
-                     embedding)
+from litellm import RateLimitError
 from litellm.llms.prompt_templates.factory import anthropic_messages_pt
 
 # litellm.num_retries=3
@@ -167,32 +163,6 @@ def test_completion_claude_3():
         pytest.fail(f"Error occurred: {e}")
 
 
-def test_completion_claude_3_empty_message():
-    litellm.set_verbose = True
-    messages = [{'role': 'user', 'content': 'please create a logo for a modern AI app. create in SVG format'},
-         {'role': 'assistant', 'content': "To create a logo for a modern AI app in SVG format, I'll use the DALL-E 3 Image Generator."},
-         {'role': 'user', 'content': 'output SVG'},
-         {'role': 'assistant', 'content': 'To generate a logo for a modern AI app in SVG format using DALL-E 3, I need to:\n1. Craft a detailed prompt describing the desired logo style and elements\n2. Specify the image size (SVG is vector-based, so size is less relevant)\n3. Call the generate_image function with the prompt and size\n4. Display the generated SVG logo using the provided syntax\nThe prompt should include keywords related to AI, modern design, and SVG format. Some key elements to incorporate could be a brain symbol, circuit lines, or a robot icon, using a minimalist style and a blue color scheme often associated with technology and intelligence.',
-          'tool_calls': [
-             {'id': 'toolu_01KEUtRVySSeMrf3g7rCA12E', 'type': 'function', 'function': {'name': 'python_tool', 'arguments': '{"code": "...python code..."}'}}
-         ]},
-         {'role': 'tool', 'content': '...python output...', 'tool_call_id': 'toolu_01KEUtRVySSeMrf3g7rCA12E'},
-         {'role': 'assistant', 'content': ''}, # empty message appended by model after tool call response
-         {'role': 'user', 'content': 'write SVG source youself!'},
-    ]
-
-    try:
-        response = completion(
-            model="anthropic/claude-3-opus-20240229",
-            messages=messages,
-            stream=True,
-            tools=[{'type': 'function', 'function': {'name': 'python_tool', 'description': 'Execute code', 'parameters': {'type': 'object', 'properties': {'headline': {'description': 'Must have. Title of this tool call (maximum 15 characters).', 'type': 'string'}, 'code': {'description': 'Python code to execute.', 'type': 'string'}}, 'required': ['code', 'headline']}}}]
-        )
-        print(response)
-    except Exception as e:
-        pytest.fail(f"Error occurred: {e}")
-
-
 def test_completion_claude_3_function_call():
     litellm.set_verbose = True
     tools = [