add anthropic scheduled tests and unit tests (#11188)

.github/workflows/scheduled_test.yml

@@ -52,6 +52,7 @@ jobs:
         shell: bash
         env:
           OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
         run: |
           make scheduled_tests
 

libs/langchain/pyproject.toml

@@ -132,6 +132,7 @@ sqlite-vss = {version = "^0.1.2", optional = true}
 anyio = "<4.0"
 jsonpatch = "^1.33"
 timescale-vector = {version = "^0.0.1", optional = true}
+anthropic = {version = "^0.3.11", optional = true}
 
 
 [tool.poetry.group.test.dependencies]
@@ -180,6 +181,7 @@ openai = "^0.27.4"
 python-dotenv = "^1.0.0"
 cassio = "^0.1.0"
 tiktoken = "^0.3.2"
+anthropic = "^0.3.11"
 
 [tool.poetry.group.lint.dependencies]
 ruff = "^0.0.249"
@@ -350,6 +352,7 @@ extended_testing = [
  "dashvector",
  "sqlite-vss",
  "timescale-vector",
+ "anthropic",
 ]
 
 [tool.ruff]

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

@@ -6,13 +6,13 @@ import pytest
 from langchain.callbacks.manager import CallbackManager
 from langchain.chat_models.anthropic import (
     ChatAnthropic,
-    convert_messages_to_prompt_anthropic,
 )
 from langchain.schema import ChatGeneration, LLMResult
 from langchain.schema.messages import AIMessage, BaseMessage, HumanMessage
 from tests.unit_tests.callbacks.fake_callback_handler import FakeCallbackHandler
 
 
+@pytest.mark.scheduled
 def test_anthropic_call() -> None:
     """Test valid call to anthropic."""
     chat = ChatAnthropic(model="test")
@@ -22,13 +22,7 @@ def test_anthropic_call() -> None:
     assert isinstance(response.content, str)
 
 
-def test_anthropic_initialization() -> None:
+@pytest.mark.scheduled
-    """Test anthropic initialization."""
-    # Verify that chat anthropic can be initialized using a secret key provided
-    # as a parameter rather than an environment variable.
-    ChatAnthropic(model="test", anthropic_api_key="test")
-
-
 def test_anthropic_generate() -> None:
     """Test generate method of anthropic."""
     chat = ChatAnthropic(model="test")
@@ -45,6 +39,7 @@ def test_anthropic_generate() -> None:
     assert chat_messages == messages_copy
 
 
+@pytest.mark.scheduled
 def test_anthropic_streaming() -> None:
     """Test streaming tokens from anthropic."""
     chat = ChatAnthropic(model="test", streaming=True)
@@ -54,6 +49,7 @@ def test_anthropic_streaming() -> None:
     assert isinstance(response.content, str)
 
 
+@pytest.mark.scheduled
 def test_anthropic_streaming_callback() -> None:
     """Test that streaming correctly invokes on_llm_new_token callback."""
     callback_handler = FakeCallbackHandler()
@@ -69,6 +65,7 @@ def test_anthropic_streaming_callback() -> None:
     assert callback_handler.llm_streams > 1
 
 
+@pytest.mark.scheduled
 @pytest.mark.asyncio
 async def test_anthropic_async_streaming_callback() -> None:
     """Test that streaming correctly invokes on_llm_new_token callback."""
@@ -90,29 +87,3 @@ async def test_anthropic_async_streaming_callback() -> None:
         assert isinstance(response, ChatGeneration)
         assert isinstance(response.text, str)
         assert response.text == response.message.content
-
-
-def test_formatting() -> None:
-    messages: List[BaseMessage] = [HumanMessage(content="Hello")]
-    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:"
-
-
-def test_anthropic_model_kwargs() -> None:
-    llm = ChatAnthropic(model_kwargs={"foo": "bar"})
-    assert llm.model_kwargs == {"foo": "bar"}
-
-
-def test_anthropic_invalid_model_kwargs() -> None:
-    with pytest.raises(ValueError):
-        ChatAnthropic(model_kwargs={"max_tokens_to_sample": 5})
-
-
-def test_anthropic_incorrect_field() -> None:
-    with pytest.warns(match="not default parameter"):
-        llm = ChatAnthropic(foo="bar")
-    assert llm.model_kwargs == {"foo": "bar"}

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

@@ -1,9 +1,12 @@
 """Test Anthropic Chat API wrapper."""
 import os
+from typing import List
 
 import pytest
 
 from langchain.chat_models import ChatAnthropic
+from langchain.chat_models.anthropic import convert_messages_to_prompt_anthropic
+from langchain.schema import AIMessage, BaseMessage, HumanMessage
 
 os.environ["ANTHROPIC_API_KEY"] = "foo"
 
@@ -37,3 +40,21 @@ def test_anthropic_incorrect_field() -> None:
     with pytest.warns(match="not default parameter"):
         llm = ChatAnthropic(foo="bar")
     assert llm.model_kwargs == {"foo": "bar"}
+
+
+@pytest.mark.requires("anthropic")
+def test_anthropic_initialization() -> None:
+    """Test anthropic initialization."""
+    # Verify that chat anthropic can be initialized using a secret key provided
+    # as a parameter rather than an environment variable.
+    ChatAnthropic(model="test", anthropic_api_key="test")
+
+
+def test_formatting() -> None:
+    messages: List[BaseMessage] = [HumanMessage(content="Hello")]
+    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:"