Anthropic system message fix (#11301)

Removes human prompt prefix before system message for anthropic models

Bedrock anthropic api enforces that Human and Assistant messages must be
interleaved (cannot have same type twice in a row). We currently treat
System Messages as human messages when converting messages -> string
prompt. Our validation when using Bedrock/BedrockChat raises an error
when this happens. For ChatAnthropic we don't validate this so no error
is raised, but perhaps the behavior is still suboptimal

libs/langchain/langchain/chat_models/anthropic.py

@@ -34,7 +34,7 @@ def _convert_one_message_to_text(
     elif isinstance(message, AIMessage):
         message_text = f"{ai_prompt} {message.content}"
     elif isinstance(message, SystemMessage):
-        message_text = f"{human_prompt} <admin>{message.content}</admin>"
+        message_text = message.content
     else:
         raise ValueError(f"Got unknown type {message}")
     return message_text
@@ -56,7 +56,6 @@ def convert_messages_to_prompt_anthropic(
     """
 
     messages = messages.copy()  # don't mutate the original list
-
     if not isinstance(messages[-1], AIMessage):
         messages.append(AIMessage(content=""))
 

libs/langchain/langchain/llms/bedrock.py

@@ -42,12 +42,12 @@ def _human_assistant_format(input_text: str) -> str:
             if count % 2 == 0:
                 count += 1
             else:
-                raise ValueError(ALTERNATION_ERROR)
+                raise ValueError(ALTERNATION_ERROR + f" Received {input_text}")
         if input_text[i : i + len(ASSISTANT_PROMPT)] == ASSISTANT_PROMPT:
             if count % 2 == 1:
                 count += 1
             else:
-                raise ValueError(ALTERNATION_ERROR)
+                raise ValueError(ALTERNATION_ERROR + f" Received {input_text}")
 
     if count % 2 == 1:  # Only saw Human, no Assistant
         input_text = input_text + ASSISTANT_PROMPT  # SILENT CORRECTION

libs/langchain/tests/unit_tests/chat_models/test_anthropic.py

@@ -6,7 +6,7 @@
 
 from langchain.chat_models import ChatAnthropic
 from langchain.chat_models.anthropic import convert_messages_to_prompt_anthropic
-from langchain.schema import AIMessage, BaseMessage, HumanMessage
+from langchain.schema import AIMessage, BaseMessage, HumanMessage, SystemMessage
 
 os.environ["ANTHROPIC_API_KEY"] = "foo"
 
@@ -50,11 +50,24 @@ def test_anthropic_initialization() -> None:
     ChatAnthropic(model="test", anthropic_api_key="test")
 
 
-def test_formatting() -> None:
-    messages: List[BaseMessage] = [HumanMessage(content="Hello")]
+@pytest.mark.parametrize(
+    ("messages", "expected"),
+    [
+        ([HumanMessage(content="Hello")], "\n\nHuman: Hello\n\nAssistant:"),
+        (
+            [HumanMessage(content="Hello"), AIMessage(content="Answer:")],
+            "\n\nHuman: Hello\n\nAssistant: Answer:",
+        ),
+        (
+            [
+                SystemMessage(content="You're an assistant"),
+                HumanMessage(content="Hello"),
+                AIMessage(content="Answer:"),
+            ],
+            "You're an assistant\n\nHuman: Hello\n\nAssistant: Answer:",
+        ),
+    ],
+)
+def test_formatting(messages: List[BaseMessage], expected: str) -> None:
     result = convert_messages_to_prompt_anthropic(messages)
-    assert result == "\n\nHuman: Hello\n\nAssistant:"
-
-    messages = [HumanMessage(content="Hello"), AIMessage(content="Answer:")]
-    result = convert_messages_to_prompt_anthropic(messages)
-    assert result == "\n\nHuman: Hello\n\nAssistant: Answer:"
+    assert result == expected