mlflow--mlflow/tests/langchain/test_langchain_model_export.py

import inspect
import json
import os
import shutil
from operator import itemgetter
from typing import Any, Iterator
from unittest import mock

import langchain
import pytest
import yaml
from langchain_community.document_loaders import TextLoader
from langchain_community.embeddings.fake import FakeEmbeddings
from langchain_community.llms import OpenAI
from langchain_community.utilities import TextRequestsWrapper
from langchain_community.vectorstores import FAISS
from langchain_core.callbacks.base import BaseCallbackHandler
from langchain_core.callbacks.manager import CallbackManagerForLLMRun
from langchain_core.language_models import SimpleChatModel
from langchain_core.messages import (
    AIMessage,
    AIMessageChunk,
    BaseMessage,
    HumanMessage,
)
from langchain_core.output_parsers import StrOutputParser
from langchain_core.outputs import ChatGenerationChunk
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder, PromptTemplate
from langchain_core.runnables import (
    RunnableBinding,
    RunnableBranch,
    RunnableLambda,
    RunnableParallel,
    RunnablePassthrough,
    RunnableSequence,
)
from langchain_core.tools import Tool
from langchain_openai import ChatOpenAI
from langchain_text_splitters.character import CharacterTextSplitter
from packaging import version
from pydantic import BaseModel
from pyspark.sql import SparkSession

import mlflow
import mlflow.models.model
import mlflow.pyfunc.scoring_server as pyfunc_scoring_server
from mlflow.deployments import PredictionsResponse
from mlflow.environment_variables import MLFLOW_CONVERT_MESSAGES_DICT_FOR_LANGCHAIN
from mlflow.exceptions import MlflowException
from mlflow.langchain.langchain_tracer import MlflowLangchainTracer
from mlflow.langchain.utils.chat import (
    try_transform_response_to_chat_format,
)
from mlflow.langchain.utils.logging import (
    IS_PICKLE_SERIALIZATION_RESTRICTED,
    lc_runnables_types,
)
from mlflow.models import Model
from mlflow.models.dependencies_schemas import DependenciesSchemasType
from mlflow.models.resources import (
    DatabricksFunction,
    DatabricksServingEndpoint,
    DatabricksSQLWarehouse,
    DatabricksVectorSearchIndex,
)
from mlflow.models.signature import Schema, infer_signature
from mlflow.models.utils import load_serving_example
from mlflow.pyfunc.context import Context
from mlflow.tracing.constant import TraceMetadataKey
from mlflow.tracing.export.inference_table import pop_trace
from mlflow.tracking.artifact_utils import _download_artifact_from_uri
from mlflow.types.schema import Array, ColSpec, DataType, Object, Property

from tests.helper_functions import _compare_logged_code_paths, pyfunc_serve_and_score_model
from tests.langchain.conftest import DeterministicDummyEmbeddings
from tests.tracing.helper import get_traces

# this kwarg was added in langchain_community 0.0.27, and
# prevents the use of pickled objects if not provided.
VECTORSTORE_KWARGS = (
    {"allow_dangerous_deserialization": True} if IS_PICKLE_SERIALIZATION_RESTRICTED else {}
)

IS_LANGCHAIN_03 = version.parse(langchain.__version__) >= version.parse("0.3.0")
IS_LANGCHAIN_v1 = version.parse(langchain.__version__).major >= 1
LANGCHAIN_V1_SKIP_REASON = "Pickle serialization is not supported for LangChain v1"

# Reusable decorator for skipping tests on LangChain v1
skip_if_v1 = pytest.mark.skipif(IS_LANGCHAIN_v1, reason=LANGCHAIN_V1_SKIP_REASON)

# langchain 0.3.30 removed legacy `.save()` from VectorStoreRetriever and chain classes,
# so the object-based `mlflow.langchain.log_model(chain, ...)` path raises NotImplementedError.
# Users on these versions should migrate to models-from-code.
skip_if_legacy_save_removed = pytest.mark.skipif(
    version.parse(langchain.__version__) >= version.parse("0.3.30"),
    reason="VectorStoreRetriever.save() removed in langchain 0.3.30+; use models-from-code",
)

# The mock OAI completion endpoint returns payload as it is
TEST_CONTENT = [{"role": "user", "content": "What is MLflow?"}]

SIMPLE_MODEL_CODE_PATH = "tests/langchain/sample_code/simple_runnable.py"


@pytest.fixture
def model_path(tmp_path):
    return tmp_path / "model"


@pytest.fixture(scope="module")
def spark():
    with SparkSession.builder.master("local[*]").getOrCreate() as s:
        yield s


def create_openai_runnable():
    from langchain_core.output_parsers import StrOutputParser

    prompt = PromptTemplate(
        input_variables=["product"],
        template="What is {product}?",
    )
    return prompt | ChatOpenAI(temperature=0.9) | StrOutputParser()


@pytest.fixture
def fake_chat_model():
    class FakeChatModel(SimpleChatModel):
        """Fake Chat Model wrapper for testing purposes."""

        endpoint_name: str = "fake-endpoint"

        def _call(
            self,
            messages: list[BaseMessage],
            stop: list[str] | None = None,
            run_manager: CallbackManagerForLLMRun | None = None,
            **kwargs: Any,
        ) -> str:
            return "Databricks"

        @property
        def _llm_type(self) -> str:
            return "fake chat model"

    return FakeChatModel(endpoint_name="fake-endpoint")


@pytest.fixture
def fake_classifier_chat_model():
    class FakeMlflowClassifier(SimpleChatModel):
        """Fake Chat Model wrapper for testing purposes."""

        def _call(
            self,
            messages: list[BaseMessage],
            stop: list[str] | None = None,
            run_manager: CallbackManagerForLLMRun | None = None,
            **kwargs: Any,
        ) -> str:
            if "MLflow" in messages[0].content.split(":")[1]:
                return "yes"
            if "cat" in messages[0].content.split(":")[1]:
                return "no"
            return "unknown"

        @property
        def _llm_type(self) -> str:
            return "fake mlflow classifier"

    return FakeMlflowClassifier()


@skip_if_v1
def test_langchain_native_log_and_load_model():
    model = create_openai_runnable()

    with mlflow.start_run():
        logged_model = mlflow.langchain.log_model(
            model, name="langchain_model", input_example={"product": "MLflow"}
        )

    loaded_model = mlflow.langchain.load_model(logged_model.model_uri)

    assert "langchain" in logged_model.flavors
    assert str(logged_model.signature.inputs) == "['product': string (required)]"
    assert str(logged_model.signature.outputs) == "[string (required)]"

    assert type(loaded_model) == RunnableSequence
    assert loaded_model.steps[0].template == "What is {product}?"
    assert type(loaded_model.steps[1]).__name__ == "ChatOpenAI"

    # Predict
    loaded_model = mlflow.pyfunc.load_model(logged_model.model_uri)
    result = loaded_model.predict([{"product": "MLflow"}])
    assert result == [json.dumps(TEST_CONTENT)]

    # Predict stream
    result = loaded_model.predict_stream([{"product": "MLflow"}])
    assert inspect.isgenerator(result)
    assert list(result) == ["Hello", " world"]


@skip_if_v1
def test_pyfunc_spark_udf_with_langchain_model(spark):
    model = create_openai_runnable()
    with mlflow.start_run():
        logged_model = mlflow.langchain.log_model(
            model, name="langchain_model", input_example={"product": "MLflow"}
        )
    loaded_model = mlflow.pyfunc.spark_udf(spark, logged_model.model_uri, result_type="string")
    df = spark.createDataFrame([("MLflow",), ("Spark",)], ["product"])
    df = df.withColumn("answer", loaded_model())
    pdf = df.toPandas()
    assert pdf["answer"].tolist() == [
        '[{"role": "user", "content": "What is MLflow?"}]',
        '[{"role": "user", "content": "What is Spark?"}]',
    ]


@pytest.mark.skipif(not IS_LANGCHAIN_v1, reason="create_agent is not supported in LangChain v0")
def test_langchain_agent_model_predict(monkeypatch):
    input_example = {"input": "What is 2 * 3?"}

    with mlflow.start_run():
        logged_model = mlflow.langchain.log_model(
            # OpenAI Client since 1.0 contains thread lock object that cannot be
            # pickled. Therefore, AgentExecutor cannot be saved with the legacy
            # object-based logging and we need to use Model-from-Code logging.
            "tests/langchain/sample_code/openai_agent.py",
            name="langchain_model",
            input_example=input_example,
        )

    loaded_model = mlflow.pyfunc.load_model(logged_model.model_uri)

    # Basic prediction
    response = loaded_model.predict([input_example])
    expected_output = "The result of 2 * 3 is 6."

    assert response[0]["messages"][-1]["content"] == expected_output

    # Stream prediction
    response = loaded_model.predict_stream([input_example])
    assert inspect.isgenerator(response)
    assert list(response) == [
        {"model": {"messages": [mock.ANY]}},
        {"tools": {"messages": [mock.ANY]}},
        {"model": {"messages": [mock.ANY]}},
    ]

    # Model serving
    inference_payload = load_serving_example(logged_model.model_uri)
    response = pyfunc_serve_and_score_model(
        logged_model.model_uri,
        data=inference_payload,
        content_type=pyfunc_scoring_server.CONTENT_TYPE_JSON,
        extra_args=["--env-manager", "local"],
    )
    # TODO: The response is not wrapped by the "predictions" key. This is a bug in
    # output handling. Often the user input contains a key "input" because it is
    # used in popular agent prompts in the hub. However, this confuses the scoring
    # server to treat it as a llm/v1/completion request.
    response = json.loads(response.content.decode("utf-8"))
    assert response[0]["messages"][-1]["content"] == expected_output


def assert_equal_retrievers(retriever, expected_retriever):
    from langchain.schema.retriever import BaseRetriever

    assert isinstance(retriever, BaseRetriever)
    assert isinstance(retriever, type(expected_retriever))
    assert isinstance(retriever.vectorstore, type(expected_retriever.vectorstore))
    assert retriever.tags == expected_retriever.tags
    assert retriever.metadata == expected_retriever.metadata
    assert retriever.search_type == expected_retriever.search_type
    assert retriever.search_kwargs == expected_retriever.search_kwargs


@skip_if_v1
@skip_if_legacy_save_removed
def test_log_and_load_retriever_chain(tmp_path):
    # Create the vector db, persist the db to a local fs folder
    loader = TextLoader("tests/langchain/state_of_the_union.txt")
    documents = loader.load()
    text_splitter = CharacterTextSplitter(chunk_size=256, chunk_overlap=0)
    docs = text_splitter.split_documents(documents)
    embeddings = DeterministicDummyEmbeddings(size=5)
    db = FAISS.from_documents(docs, embeddings)
    persist_dir = str(tmp_path / "faiss_index")
    db.save_local(persist_dir)

    # Define the loader_fn
    def load_retriever(persist_directory):
        import numpy as np
        from langchain.embeddings.base import Embeddings

        class DeterministicDummyEmbeddings(Embeddings, BaseModel):
            size: int

            def _get_embedding(self, text: str) -> list[float]:
                if isinstance(text, np.ndarray):
                    text = text.item()
                seed = abs(hash(text)) % (10**8)
                np.random.seed(seed)
                return list(np.random.normal(size=self.size))

            def embed_documents(self, texts: list[str]) -> list[list[float]]:
                return [self._get_embedding(t) for t in texts]

            def embed_query(self, text: str) -> list[float]:
                return self._get_embedding(text)

        embeddings = DeterministicDummyEmbeddings(size=5)
        vectorstore = FAISS.load_local(
            persist_directory,
            embeddings,
            **VECTORSTORE_KWARGS,
        )
        return vectorstore.as_retriever()

    query = "What did the president say about Ketanji Brown Jackson"
    langchain_input = {"query": query}
    # Log the retriever
    with mlflow.start_run():
        logged_model = mlflow.langchain.log_model(
            db.as_retriever(),
            name="retriever",
            loader_fn=load_retriever,
            persist_dir=persist_dir,
            input_example=langchain_input,
        )

    # Remove the persist_dir
    shutil.rmtree(persist_dir)

    # Load the retriever
    loaded_model = mlflow.langchain.load_model(logged_model.model_uri)
    assert_equal_retrievers(loaded_model, db.as_retriever())

    loaded_pyfunc_model = mlflow.pyfunc.load_model(logged_model.model_uri)
    result = loaded_pyfunc_model.predict([langchain_input])
    expected_result = [
        {
            "page_content": doc.page_content,
            "metadata": doc.metadata,
            "type": "Document",
            "id": mock.ANY,
        }
        for doc in db.as_retriever().get_relevant_documents(query)
    ]
    assert result == [expected_result]

    # Serve the retriever
    inference_payload = load_serving_example(logged_model.model_uri)
    response = pyfunc_serve_and_score_model(
        logged_model.model_uri,
        data=inference_payload,
        content_type=pyfunc_scoring_server.CONTENT_TYPE_JSON,
        extra_args=["--env-manager", "local"],
    )
    pred = PredictionsResponse.from_json(response.content.decode("utf-8"))["predictions"]
    assert type(pred) == list
    assert len(pred) == 1
    docs_list = pred[0]
    assert type(docs_list) == list
    assert len(docs_list) == 4
    # The returned docs are non-deterministic when used with dummy embeddings,
    # so we cannot assert pred == {"predictions": [expected_result]}


def load_requests_wrapper(_):
    return TextRequestsWrapper(headers=None, aiosession=None)


@skip_if_v1
@skip_if_legacy_save_removed
def test_agent_with_unpicklable_tools(tmp_path):
    from langchain.agents import AgentType, initialize_agent

    tmp_file = tmp_path / "temp_file.txt"
    with open(tmp_file, mode="w") as temp_file:
        # files that aren't opened for reading cannot be pickled
        tools = [
            Tool.from_function(
                func=lambda: temp_file,
                name="Write 0",
                description="If you need to write 0 to a file",
            )
        ]
        agent = initialize_agent(
            llm=OpenAI(temperature=0),
            tools=tools,
            agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION,
        )

        with pytest.raises(
            MlflowException,
            match=(
                "Error when attempting to pickle the AgentExecutor tools. "
                "This model likely does not support serialization."
            ),
        ):
            with mlflow.start_run():
                mlflow.langchain.log_model(agent, name="unpicklable_tools")


@skip_if_v1
def test_save_load_runnable_passthrough():
    runnable = RunnablePassthrough()
    assert runnable.invoke("hello") == "hello"

    input_example = "hello"
    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            runnable, name="model_path", input_example=input_example
        )

    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    assert loaded_model.invoke(input_example) == "hello"
    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    assert pyfunc_loaded_model.predict(["hello"]) == ["hello"]


@skip_if_v1
def test_save_load_runnable_lambda(spark):
    def add_one(x: int) -> int:
        return x + 1

    runnable = RunnableLambda(add_one)

    assert runnable.invoke(1) == 2
    assert runnable.batch([1, 2, 3]) == [2, 3, 4]

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            runnable, name="runnable_lambda", input_example=[1, 2, 3]
        )

    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    assert loaded_model.invoke(1) == 2
    assert loaded_model.batch([1, 2, 3]) == [2, 3, 4]

    loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    assert loaded_model.predict(1) == [2]
    assert loaded_model.predict([1, 2, 3]) == [2, 3, 4]

    udf = mlflow.pyfunc.spark_udf(spark, model_info.model_uri, result_type="long")
    df = spark.createDataFrame([(1,), (2,), (3,)], ["data"])
    df = df.withColumn("answer", udf("data"))
    pdf = df.toPandas()
    assert pdf["answer"].tolist() == [2, 3, 4]


@skip_if_v1
def test_save_load_runnable_lambda_in_sequence():
    def add_one(x):
        return x + 1

    def mul_two(x):
        return x * 2

    runnable_1 = RunnableLambda(add_one)
    runnable_2 = RunnableLambda(mul_two)
    sequence = runnable_1 | runnable_2
    assert sequence.invoke(1) == 4

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            sequence, name="model_path", input_example=[1, 2, 3]
        )

    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    assert loaded_model.invoke(1) == 4
    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    assert pyfunc_loaded_model.predict(1) == [4]
    assert pyfunc_loaded_model.predict([1, 2, 3]) == [4, 6, 8]

    inference_payload = load_serving_example(model_info.model_uri)
    response = pyfunc_serve_and_score_model(
        model_info.model_uri,
        data=inference_payload,
        content_type=pyfunc_scoring_server.CONTENT_TYPE_JSON,
        extra_args=["--env-manager", "local"],
    )
    assert PredictionsResponse.from_json(response.content.decode("utf-8")) == {
        "predictions": [4, 6, 8]
    }


@skip_if_v1
def test_predict_with_callbacks(fake_chat_model):
    class TestCallbackHandler(BaseCallbackHandler):
        def __init__(self):
            super().__init__()
            self.num_llm_start_calls = 0

        def on_llm_start(
            self,
            serialized: dict[str, Any],
            prompts: list[str],
            **kwargs: Any,
        ) -> Any:
            self.num_llm_start_calls += 1

    prompt = ChatPromptTemplate.from_template("What's your favorite {industry} company?")
    chain = prompt | fake_chat_model | StrOutputParser()
    # Test the basic functionality of the chain
    assert chain.invoke({"industry": "tech"}) == "Databricks"

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            chain, name="model_path", input_example={"industry": "tech"}
        )

    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)

    callback_handler1 = TestCallbackHandler()
    callback_handler2 = TestCallbackHandler()

    # Ensure handlers have not been called yet
    assert callback_handler1.num_llm_start_calls == 0
    assert callback_handler2.num_llm_start_calls == 0

    assert (
        pyfunc_loaded_model._model_impl._predict_with_callbacks(
            {"industry": "tech"},
            callback_handlers=[callback_handler1, callback_handler2],
        )
        == "Databricks"
    )

    # Test that the callback handlers were called
    assert callback_handler1.num_llm_start_calls == 1
    assert callback_handler2.num_llm_start_calls == 1

    inference_payload = load_serving_example(model_info.model_uri)
    response = pyfunc_serve_and_score_model(
        model_info.model_uri,
        data=inference_payload,
        content_type=pyfunc_scoring_server.CONTENT_TYPE_JSON,
        extra_args=["--env-manager", "local"],
    )
    assert PredictionsResponse.from_json(response.content.decode("utf-8")) == {
        "predictions": ["Databricks"]
    }


@skip_if_v1
def test_predict_with_callbacks_supports_chat_response_conversion(fake_chat_model):
    prompt = ChatPromptTemplate.from_template("What's your favorite {industry} company?")
    chain = prompt | fake_chat_model | StrOutputParser()
    # Test the basic functionality of the chain
    assert chain.invoke({"industry": "tech"}) == "Databricks"

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            chain, name="model_path", input_example={"industry": "tech"}
        )

    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    expected_chat_response = {
        "id": None,
        "object": "chat.completion",
        "created": 1677858242,
        "model": "",
        "choices": [
            {
                "index": 0,
                "message": {
                    "role": "assistant",
                    "content": "Databricks",
                },
                "finish_reason": None,
            }
        ],
        "usage": {
            "prompt_tokens": None,
            "completion_tokens": None,
            "total_tokens": None,
        },
    }
    with mock.patch("time.time", return_value=1677858242):
        assert (
            pyfunc_loaded_model._model_impl._predict_with_callbacks(
                {"industry": "tech"},
                convert_chat_responses=True,
            )
            == expected_chat_response
        )

        assert (
            pyfunc_loaded_model._model_impl._predict_with_callbacks(
                {"industry": "tech"},
                convert_chat_responses=False,
            )
            == "Databricks"
        )


@skip_if_v1
def test_save_load_runnable_parallel():
    runnable = RunnableParallel({"llm": create_openai_runnable()})
    expected_result = {"llm": json.dumps(TEST_CONTENT)}
    assert runnable.invoke({"product": "MLflow"}) == expected_result
    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            runnable, name="model_path", input_example=[{"product": "MLflow"}]
        )
    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    assert loaded_model.invoke({"product": "MLflow"}) == expected_result
    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    assert pyfunc_loaded_model.predict([{"product": "MLflow"}]) == [expected_result]

    inference_payload = load_serving_example(model_info.model_uri)
    response = pyfunc_serve_and_score_model(
        model_info.model_uri,
        data=inference_payload,
        content_type=pyfunc_scoring_server.CONTENT_TYPE_JSON,
        extra_args=["--env-manager", "local"],
    )
    assert PredictionsResponse.from_json(response.content.decode("utf-8")) == {
        "predictions": [expected_result]
    }


@skip_if_v1
def test_save_load_chain_with_model_paths():
    prompt1 = PromptTemplate.from_template("what is the city {person} is from?")
    llm = ChatOpenAI(temperature=0.9)
    model = prompt1 | llm | StrOutputParser()

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(model, name="model_path")
    artifact_path = "model_path"
    with (
        mlflow.start_run(),
        mock.patch("mlflow.langchain.model._add_code_from_conf_to_system_path") as add_mock,
    ):
        model_info = mlflow.langchain.log_model(model, name=artifact_path, code_paths=[__file__])
        mlflow.langchain.load_model(model_info.model_uri)
        model_uri = model_info.model_uri
        _compare_logged_code_paths(__file__, model_uri, mlflow.langchain.FLAVOR_NAME)
        add_mock.assert_called()


@skip_if_v1
@skip_if_legacy_save_removed
def test_save_load_rag(tmp_path, spark, fake_chat_model):
    # TODO: Migrate to models-from-code
    # Create the vector db, persist the db to a local fs folder
    loader = TextLoader("tests/langchain/state_of_the_union.txt")
    documents = loader.load()
    text_splitter = CharacterTextSplitter(chunk_size=10, chunk_overlap=0)
    docs = text_splitter.split_documents(documents)
    embeddings = DeterministicDummyEmbeddings(size=5)
    db = FAISS.from_documents(docs, embeddings)
    persist_dir = str(tmp_path / "faiss_index")
    db.save_local(persist_dir)
    retriever = db.as_retriever()

    def load_retriever(persist_directory):
        embeddings = FakeEmbeddings(size=5)
        vectorstore = FAISS.load_local(
            persist_directory,
            embeddings,
            **VECTORSTORE_KWARGS,
        )
        return vectorstore.as_retriever()

    prompt = ChatPromptTemplate.from_template(
        "Answer the following question based on the context: {context}\nQuestion: {question}"
    )
    retrieval_chain = (
        {
            "context": retriever,
            "question": RunnablePassthrough(),
        }
        | prompt
        | fake_chat_model
        | StrOutputParser()
    )
    question = "What is a good name for a company that makes MLflow?"
    answer = "Databricks"
    assert retrieval_chain.invoke(question) == answer
    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            retrieval_chain,
            name="model_path",
            loader_fn=load_retriever,
            persist_dir=persist_dir,
            input_example=question,
        )

    # Remove the persist_dir
    shutil.rmtree(persist_dir)

    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    assert loaded_model.invoke(question) == answer
    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    assert pyfunc_loaded_model.predict(question) == [answer]

    udf = mlflow.pyfunc.spark_udf(spark, model_info.model_uri, result_type="string")
    df = spark.createDataFrame([(question,), (question,)], ["question"])
    df = df.withColumn("answer", udf("question"))
    pdf = df.toPandas()
    assert pdf["answer"].tolist() == [answer, answer]

    inference_payload = load_serving_example(model_info.model_uri)
    response = pyfunc_serve_and_score_model(
        model_info.model_uri,
        data=inference_payload,
        content_type=pyfunc_scoring_server.CONTENT_TYPE_JSON,
        extra_args=["--env-manager", "local"],
    )
    assert PredictionsResponse.from_json(response.content.decode("utf-8")) == {
        "predictions": [answer]
    }


@skip_if_v1
def test_runnable_branch_save_load():
    branch = RunnableBranch(
        (lambda x: isinstance(x, str), lambda x: x.upper()),
        (lambda x: isinstance(x, int), lambda x: x + 1),
        (lambda x: isinstance(x, float), lambda x: x * 2),
        lambda x: "goodbye",
    )

    assert branch.invoke("hello") == "HELLO"
    assert branch.invoke({}) == "goodbye"

    with mlflow.start_run():
        # We only support single input format for now, so we should
        # not save signature for runnable branch which accepts multiple
        # input types
        model_info = mlflow.langchain.log_model(branch, name="model_path")

    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    assert loaded_model.invoke("hello") == "HELLO"
    assert loaded_model.invoke({}) == "goodbye"
    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    assert pyfunc_loaded_model.predict("hello") == "HELLO"
    assert pyfunc_loaded_model.predict({}) == "goodbye"

    response = pyfunc_serve_and_score_model(
        model_info.model_uri,
        data=json.dumps({"inputs": "hello"}),
        content_type=pyfunc_scoring_server.CONTENT_TYPE_JSON,
        extra_args=["--env-manager", "local"],
    )
    assert PredictionsResponse.from_json(response.content.decode("utf-8")) == {
        "predictions": "HELLO"
    }


@skip_if_v1
def test_complex_runnable_branch_save_load(fake_chat_model, fake_classifier_chat_model):
    prompt = ChatPromptTemplate.from_template("{question_is_relevant}\n{query}")
    # Need to add prompt here as the chat model doesn't accept dict input
    answer_model = prompt | fake_chat_model

    decline_to_answer = RunnableLambda(
        lambda x: "I cannot answer questions that are not about MLflow."
    )
    something_went_wrong = RunnableLambda(lambda x: "Something went wrong.")

    is_question_about_mlflow_prompt = ChatPromptTemplate.from_template(
        "You are classifying documents to know if this question "
        "is related with MLflow. Only answer with yes or no. The question is: {query}"
    )

    branch_node = RunnableBranch(
        (lambda x: x["question_is_relevant"].lower() == "yes", answer_model),
        (lambda x: x["question_is_relevant"].lower() == "no", decline_to_answer),
        something_went_wrong,
    )

    chain = (
        {
            "question_is_relevant": is_question_about_mlflow_prompt
            | fake_classifier_chat_model
            | StrOutputParser(),
            "query": itemgetter("query"),
        }
        | branch_node
        | StrOutputParser()
    )

    assert chain.invoke({"query": "Who owns MLflow?"}) == "Databricks"
    assert (
        chain.invoke({"query": "Do you like cat?"})
        == "I cannot answer questions that are not about MLflow."
    )
    assert chain.invoke({"query": "Are you happy today?"}) == "Something went wrong."

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            chain, name="model_path", input_example={"query": "Who owns MLflow?"}
        )

    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    assert loaded_model.invoke({"query": "Who owns MLflow?"}) == "Databricks"
    assert (
        loaded_model.invoke({"query": "Do you like cat?"})
        == "I cannot answer questions that are not about MLflow."
    )
    assert loaded_model.invoke({"query": "Are you happy today?"}) == "Something went wrong."
    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    assert pyfunc_loaded_model.predict({"query": "Who owns MLflow?"}) == ["Databricks"]
    assert pyfunc_loaded_model.predict({"query": "Do you like cat?"}) == [
        "I cannot answer questions that are not about MLflow."
    ]
    assert pyfunc_loaded_model.predict({"query": "Are you happy today?"}) == [
        "Something went wrong."
    ]

    inference_payload = load_serving_example(model_info.model_uri)
    response = pyfunc_serve_and_score_model(
        model_info.model_uri,
        data=inference_payload,
        content_type=pyfunc_scoring_server.CONTENT_TYPE_JSON,
        extra_args=["--env-manager", "local"],
    )
    assert PredictionsResponse.from_json(response.content.decode("utf-8")) == {
        "predictions": ["Databricks"]
    }


@skip_if_v1
def test_chat_with_history(spark, fake_chat_model):
    prompt_with_history_str = """
    Here is a history between you and a human: {chat_history}

    Now, please answer this question: {question}
    """

    prompt_with_history = PromptTemplate(
        input_variables=["chat_history", "question"], template=prompt_with_history_str
    )

    def extract_question(input):
        return input[-1]["content"]

    def extract_history(input):
        return input[:-1]

    chain_with_history = (
        {
            "question": itemgetter("messages") | RunnableLambda(extract_question),
            "chat_history": itemgetter("messages") | RunnableLambda(extract_history),
        }
        | prompt_with_history
        | fake_chat_model
        | StrOutputParser()
    )

    input_example = {"messages": [{"role": "user", "content": "Who owns MLflow?"}]}
    assert chain_with_history.invoke(input_example) == "Databricks"

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            chain_with_history, name="model_path", input_example=input_example
        )
    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    assert loaded_model.invoke(input_example) == "Databricks"
    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    input_schema = pyfunc_loaded_model.metadata.get_input_schema()
    assert input_schema == Schema([
        ColSpec(
            Array(
                Object([
                    Property("role", DataType.string),
                    Property("content", DataType.string),
                ])
            ),
            "messages",
        )
    ])
    assert pyfunc_loaded_model.predict(input_example) == ["Databricks"]

    udf = mlflow.pyfunc.spark_udf(spark, model_info.model_uri, result_type="string")
    df = spark.createDataFrame([(input_example["messages"],)], ["messages"])
    df = df.withColumn("answer", udf("messages"))
    pdf = df.toPandas()
    assert pdf["answer"].tolist() == ["Databricks"]

    inference_payload = load_serving_example(model_info.model_uri)
    response = pyfunc_serve_and_score_model(
        model_info.model_uri,
        data=inference_payload,
        content_type=pyfunc_scoring_server.CONTENT_TYPE_JSON,
        extra_args=["--env-manager", "local"],
    )
    assert json.loads(response.content.decode("utf-8")) == ["Databricks"]


class ChatModel(SimpleChatModel):
    def _call(self, messages, stop, run_manager, **kwargs):
        return "\n".join([f"{message.type}: {message.content}" for message in messages])

    @property
    def _llm_type(self) -> str:
        return "chat model"


@skip_if_v1
def test_predict_with_builtin_pyfunc_chat_conversion(spark):
    # TODO: Migrate to models-from-code
    input_example = {
        "messages": [
            {"role": "system", "content": "You are a helpful assistant."},
            {"role": "assistant", "content": "What would you like to ask?"},
            {"role": "user", "content": "Who owns MLflow?"},
        ]
    }
    content = (
        "system: You are a helpful assistant.\n"
        "ai: What would you like to ask?\n"
        "human: Who owns MLflow?"
    )

    chain = ChatModel() | StrOutputParser()
    assert chain.invoke([HumanMessage(content="Who owns MLflow?")]) == "human: Who owns MLflow?"
    with pytest.raises(ValueError, match="Invalid input type"):
        chain.invoke(input_example)

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            chain, name="model_path", input_example=input_example
        )

    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    assert (
        loaded_model.invoke([HumanMessage(content="Who owns MLflow?")]) == "human: Who owns MLflow?"
    )

    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    expected_chat_response = {
        "id": None,
        "object": "chat.completion",
        "created": 1677858242,
        "model": "",
        "choices": [
            {
                "index": 0,
                "message": {
                    "role": "assistant",
                    "content": content,
                },
                "finish_reason": None,
            }
        ],
        "usage": {
            "prompt_tokens": None,
            "completion_tokens": None,
            "total_tokens": None,
        },
    }

    with mock.patch("time.time", return_value=1677858242):
        result1 = pyfunc_loaded_model.predict(input_example)
        result1[0]["id"] = None
        assert result1 == [expected_chat_response]
        result2 = pyfunc_loaded_model.predict([input_example, input_example])
        result2[0]["id"] = None
        result2[1]["id"] = None
        assert result2 == [
            expected_chat_response,
            expected_chat_response,
        ]

    with pytest.raises(MlflowException, match="Unrecognized chat message role"):
        pyfunc_loaded_model.predict({"messages": [{"role": "foobar", "content": "test content"}]})


@skip_if_v1
def test_predict_with_builtin_pyfunc_chat_conversion_for_aimessage_response():
    class ChatModel(SimpleChatModel):
        def _call(self, messages, stop, run_manager, **kwargs):
            return "You own MLflow"

        @property
        def _llm_type(self) -> str:
            return "chat model"

    input_example = {
        "messages": [
            {"role": "system", "content": "You are a helpful assistant."},
            {"role": "assistant", "content": "What would you like to ask?"},
            {"role": "user", "content": "Who owns MLflow?"},
        ]
    }

    chain = ChatModel()
    result = chain.invoke([HumanMessage(content="Who owns MLflow?")])
    assert isinstance(result, AIMessage)
    assert result.content == "You own MLflow"

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            chain, name="model_path", input_example=input_example
        )

    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    result = loaded_model.invoke([HumanMessage(content="Who owns MLflow?")])
    assert isinstance(result, AIMessage)
    assert result.content == "You own MLflow"

    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    with mock.patch("time.time", return_value=1677858242):
        result = pyfunc_loaded_model.predict(input_example)
        assert "id" in result[0], "Response message id is lost."
        result[0]["id"] = None
        assert result == [
            {
                "id": None,
                "object": "chat.completion",
                "created": 1677858242,
                "model": "",
                "choices": [
                    {
                        "index": 0,
                        "message": {
                            "role": "assistant",
                            "content": "You own MLflow",
                        },
                        "finish_reason": None,
                    }
                ],
                "usage": {
                    "prompt_tokens": None,
                    "completion_tokens": None,
                    "total_tokens": None,
                },
            }
        ]


@skip_if_v1
def test_pyfunc_builtin_chat_request_conversion_fails_gracefully():
    chain = RunnablePassthrough() | itemgetter("messages")
    # Ensure we're going to test that "messages" remains intact & unchanged even if it
    # doesn't appear explicitly in the chain's input schema
    assert "messages" not in chain.input_schema().model_fields

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(chain, name="model_path")
        pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)

    assert pyfunc_loaded_model.predict({"messages": "not an array"}) == "not an array"

    # Verify that messages aren't converted to LangChain format if extra keys are present,
    # under the assumption that additional keys can't be specified when calling LangChain invoke()
    # / batch() with chat messages
    assert pyfunc_loaded_model.predict({
        "messages": [{"role": "user", "content": "blah"}],
        "extrakey": "extra",
    }) == [
        {"role": "user", "content": "blah"},
    ]

    # Verify that messages aren't converted to LangChain format if role / content are missing
    # or extra keys are present in the message
    assert pyfunc_loaded_model.predict({
        "messages": [{"content": "blah"}],
    }) == [
        {"content": "blah"},
    ]
    assert pyfunc_loaded_model.predict({
        "messages": [{"role": "user", "content": "blah"}, {}],
    }) == [
        {"role": "user", "content": "blah"},
        {},
    ]
    assert pyfunc_loaded_model.predict({
        "messages": [{"role": "user", "content": 123}],
    }) == [
        {"role": "user", "content": 123},
    ]

    # Verify behavior for batches of message histories
    assert pyfunc_loaded_model.predict([
        {
            "messages": "not an array",
        },
        {
            "messages": [{"role": "user", "content": "content"}],
        },
    ]) == [
        "not an array",
        [{"role": "user", "content": "content"}],
    ]
    assert pyfunc_loaded_model.predict([
        {
            "messages": [{"role": "user", "content": "content"}],
        },
        {"messages": [{"role": "user", "content": "content"}], "extrakey": "extra"},
    ]) == [
        [{"role": "user", "content": "content"}],
        [{"role": "user", "content": "content"}],
    ]
    assert pyfunc_loaded_model.predict([
        {
            "messages": [{"role": "user", "content": "content"}],
        },
        {
            "messages": [
                {"role": "user", "content": "content"},
                {"role": "user", "content": 123},
            ],
        },
    ]) == [
        [{"role": "user", "content": "content"}],
        [{"role": "user", "content": "content"}, {"role": "user", "content": 123}],
    ]


@skip_if_v1
def test_save_load_chain_that_relies_on_pickle_serialization(monkeypatch, model_path):
    from langchain_community.llms.databricks import Databricks

    monkeypatch.setattr(
        "langchain_community.llms.databricks._DatabricksServingEndpointClient",
        mock.MagicMock(),
    )
    monkeypatch.setenv("DATABRICKS_HOST", "test-host")
    monkeypatch.setenv("DATABRICKS_TOKEN", "test-token")

    llm_kwargs = {"endpoint_name": "test-endpoint", "temperature": 0.9}
    if IS_PICKLE_SERIALIZATION_RESTRICTED:
        llm_kwargs["allow_dangerous_deserialization"] = True

    llm = Databricks(**llm_kwargs)
    prompt = PromptTemplate(input_variables=["question"], template="I have a question: {question}")
    chain = prompt | llm | StrOutputParser()

    # Not passing an input_example to avoid triggering prediction
    mlflow.langchain.save_model(chain, model_path)

    loaded_model = mlflow.langchain.load_model(model_path)

    # Check if the deserialized model has the same endpoint and temperature
    loaded_databricks_llm = loaded_model.middle[0]
    assert loaded_databricks_llm.endpoint_name == "test-endpoint"
    assert loaded_databricks_llm.temperature == 0.9


def _get_message_content(predictions):
    return predictions[0]["choices"][0]["message"]["content"]


@pytest.mark.parametrize(
    ("chain_path", "model_config"),
    [
        (
            os.path.abspath("tests/langchain/sample_code/chain.py"),
            os.path.abspath("tests/langchain/sample_code/config.yml"),
        ),
        (
            "tests/langchain/../langchain/sample_code/chain.py",
            "tests/langchain/../langchain/sample_code/config.yml",
        ),
    ],
)
def test_save_load_chain_as_code(chain_path, model_config, monkeypatch):
    input_example = {
        "messages": [
            {
                "role": "user",
                "content": "What is a good name for a company that makes MLflow?",
            }
        ]
    }
    artifact_path = "model_path"
    with mlflow.start_run() as run:
        model_info = mlflow.langchain.log_model(
            chain_path,
            name=artifact_path,
            input_example=input_example,
            model_config=model_config,
        )

    client = mlflow.tracking.MlflowClient()
    run_id = run.info.run_id
    assert client.get_run(run_id).data.params == {
        "llm_prompt_template": "Answer the following question based on "
        "the context: {context}\nQuestion: {question}",
        "embedding_size": "5",
        "not_used_array": "[1, 2, 3]",
        "response": "Databricks",
    }

    assert mlflow.models.model_config.__mlflow_model_config__ is None
    loaded_model = mlflow.langchain.load_model(model_info.model_uri)

    # During the loading process, MLflow executes the chain.py file to
    # load the model class. It should not generate any traces even if
    # the code enables autologging and invoke chain.
    assert len(get_traces()) == 0

    assert mlflow.models.model_config.__mlflow_model_config__ is None
    answer = "Databricks"
    assert loaded_model.invoke(input_example) == answer
    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    assert answer == _get_message_content(pyfunc_loaded_model.predict(input_example))

    inference_payload = load_serving_example(model_info.model_uri)
    response = pyfunc_serve_and_score_model(
        model_info.model_uri,
        data=inference_payload,
        content_type=pyfunc_scoring_server.CONTENT_TYPE_JSON,
        extra_args=["--env-manager", "local"],
    )
    predictions = json.loads(response.content.decode("utf-8"))
    # Mock out the `created` timestamp as it is not deterministic
    expected = [{**try_transform_response_to_chat_format(answer), "created": mock.ANY}]
    assert expected == predictions

    pyfunc_model_path = _download_artifact_from_uri(model_info.model_uri)
    reloaded_model = Model.load(os.path.join(pyfunc_model_path, "MLmodel"))
    assert reloaded_model.resources["databricks"] == {
        "serving_endpoint": [{"name": "fake-endpoint"}]
    }
    assert reloaded_model.metadata["dependencies_schemas"] == {
        DependenciesSchemasType.RETRIEVERS.value: [
            {
                "doc_uri": "doc-uri",
                "name": "retriever",
                "other_columns": ["column1", "column2"],
                "primary_key": "primary-key",
                "text_column": "text-column",
            }
        ]
    }

    # Emulate the model serving environment
    monkeypatch.setenv("IS_IN_DB_MODEL_SERVING_ENV", "true")
    monkeypatch.setenv("ENABLE_MLFLOW_TRACING", "true")
    mlflow.tracing.reset()

    request_id = "mock_request_id"
    tracer = MlflowLangchainTracer(prediction_context=Context(request_id))
    input_example = {"messages": [{"role": "user", "content": json.dumps(TEST_CONTENT)}]}
    response = pyfunc_loaded_model._model_impl._predict_with_callbacks(
        data=input_example, callback_handlers=[tracer]
    )
    assert response["choices"][0]["message"]["content"] == "Databricks"
    trace = pop_trace(request_id)
    assert trace["info"]["tags"][DependenciesSchemasType.RETRIEVERS.value] == json.dumps([
        {
            "doc_uri": "doc-uri",
            "name": "retriever",
            "other_columns": ["column1", "column2"],
            "primary_key": "primary-key",
            "text_column": "text-column",
        }
    ])


@pytest.mark.parametrize(
    "chain_path",
    [
        os.path.abspath("tests/langchain/sample_code/chain.py"),
        "tests/langchain/../langchain/sample_code/chain.py",
    ],
)
def test_save_load_chain_as_code_model_config_dict(chain_path):
    input_example = {
        "messages": [
            {
                "role": "user",
                "content": "What is a good name for a company that makes MLflow?",
            }
        ]
    }
    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            chain_path,
            name="model_path",
            input_example=input_example,
            model_config={
                "response": "modified response",
                "embedding_size": 5,
                "llm_prompt_template": "answer the question",
            },
        )

    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    answer = "modified response"
    assert loaded_model.invoke(input_example) == answer
    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    assert answer == _get_message_content(pyfunc_loaded_model.predict(input_example))


@pytest.mark.parametrize(
    "model_config",
    [
        os.path.abspath("tests/langchain/sample_code/config.yml"),
        "tests/langchain/../langchain/sample_code/config.yml",
    ],
)
def test_save_load_chain_as_code_with_different_names(tmp_path, model_config):
    input_example = {
        "messages": [
            {
                "role": "user",
                "content": "What is a good name for a company that makes MLflow?",
            }
        ]
    }

    # Read the contents of the original chain file
    with open("tests/langchain/sample_code/chain.py") as chain_file:
        chain_file_content = chain_file.read()

    temp_file = tmp_path / "model.py"
    temp_file.write_text(chain_file_content)

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            str(temp_file),
            name="model_path",
            input_example=input_example,
            model_config=model_config,
        )

    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    answer = "Databricks"
    assert loaded_model.invoke(input_example) == answer
    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    assert answer == _get_message_content(pyfunc_loaded_model.predict(input_example))


@pytest.mark.parametrize(
    "chain_path",
    [
        os.path.abspath("tests/langchain/sample_code/chain.py"),
        "tests/langchain/../langchain/sample_code/chain.py",
    ],
)
@pytest.mark.parametrize(
    "model_config",
    [
        os.path.abspath("tests/langchain/sample_code/config.yml"),
        "tests/langchain/../langchain/sample_code/config.yml",
    ],
)
def test_save_load_chain_as_code_multiple_times(tmp_path, chain_path, model_config):
    input_example = {
        "messages": [
            {
                "role": "user",
                "content": "What is a good name for a company that makes MLflow?",
            }
        ]
    }
    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            chain_path,
            name="model_path",
            input_example=input_example,
            model_config=model_config,
        )

    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    with open(model_config) as f:
        base_config = yaml.safe_load(f)

    assert loaded_model.middle[0].messages[0].prompt.template == base_config["llm_prompt_template"]

    file_name = "config_updated.yml"
    new_config_file = str(tmp_path.joinpath(file_name))

    new_config = base_config.copy()
    new_config["llm_prompt_template"] = "new_template"
    with open(new_config_file, "w") as f:
        yaml.dump(new_config, f)

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            chain_path,
            name="model_path",
            input_example=input_example,
            model_config=new_config_file,
        )

    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    assert loaded_model.middle[0].messages[0].prompt.template == new_config["llm_prompt_template"]


@pytest.mark.parametrize(
    "chain_path",
    [
        os.path.abspath("tests/langchain/sample_code/chain.py"),
        "tests/langchain/../langchain/sample_code/chain.py",
    ],
)
def test_save_load_chain_as_code_with_model_paths(chain_path):
    input_example = {
        "messages": [
            {
                "role": "user",
                "content": "What is a good name for a company that makes MLflow?",
            }
        ]
    }
    artifact_path = "model_path"
    with (
        mlflow.start_run(),
        mock.patch("mlflow.langchain.model._add_code_from_conf_to_system_path") as add_mock,
    ):
        model_info = mlflow.langchain.log_model(
            chain_path,
            name=artifact_path,
            input_example=input_example,
            code_paths=[__file__],
            model_config={
                "response": "modified response",
                "embedding_size": 5,
                "llm_prompt_template": "answer the question",
            },
        )
        loaded_model = mlflow.langchain.load_model(model_info.model_uri)
        answer = "modified response"
        _compare_logged_code_paths(__file__, model_info.model_uri, mlflow.langchain.FLAVOR_NAME)
        assert loaded_model.invoke(input_example) == answer
        add_mock.assert_called()


@pytest.mark.parametrize("chain_path", [os.path.abspath("tests/langchain1/sample_code/chain.py")])
def test_save_load_chain_errors(chain_path):
    input_example = {
        "messages": [
            {
                "role": "user",
                "content": "What is a good name for a company that makes MLflow?",
            }
        ]
    }
    with mlflow.start_run():
        with pytest.raises(
            MlflowException,
            match=f"The provided model path '{chain_path}' does not exist. "
            "Ensure the file path is valid and try again.",
        ):
            mlflow.langchain.log_model(
                chain_path,
                name="model_path",
                input_example=input_example,
                model_config="tests/langchain/state_of_the_union.txt",
            )


@pytest.mark.parametrize(
    "chain_path",
    [
        os.path.abspath("tests/langchain/sample_code/no_config/chain.py"),
        "tests/langchain/../langchain/sample_code/no_config/chain.py",
    ],
)
def test_save_load_chain_as_code_optional_code_path(chain_path):
    input_example = {
        "messages": [
            {
                "role": "user",
                "content": "What is a good name for a company that makes MLflow?",
            }
        ]
    }
    artifact_path = "new_model_path"
    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            chain_path,
            name=artifact_path,
            input_example=input_example,
        )

    assert mlflow.models.model_config.__mlflow_model_config__ is None
    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    assert mlflow.models.model_config.__mlflow_model_config__ is None
    answer = "Databricks"
    assert loaded_model.invoke(input_example) == answer
    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    assert (
        pyfunc_loaded_model
        .predict(input_example)[0]
        .get("choices")[0]
        .get("message")
        .get("content")
        == answer
    )

    inference_payload = load_serving_example(model_info.model_uri)
    response = pyfunc_serve_and_score_model(
        model_info.model_uri,
        data=inference_payload,
        content_type=pyfunc_scoring_server.CONTENT_TYPE_JSON,
        extra_args=["--env-manager", "local"],
    )
    # avoid minor diff of created time in the response
    prediction_result = json.loads(response.content.decode("utf-8"))
    prediction_result[0]["created"] = 123
    expected_prediction = try_transform_response_to_chat_format(answer)
    expected_prediction["created"] = 123
    assert prediction_result == [expected_prediction]

    pyfunc_model_path = _download_artifact_from_uri(model_info.model_uri)
    reloaded_model = Model.load(os.path.join(pyfunc_model_path, "MLmodel"))
    assert reloaded_model.resources["databricks"] == {
        "serving_endpoint": [{"name": "fake-endpoint"}]
    }
    assert reloaded_model.metadata is None


@pytest.fixture
def fake_chat_stream_model():
    class FakeChatStreamModel(SimpleChatModel):
        """Fake Chat Stream Model wrapper for testing purposes."""

        endpoint_name: str = "fake-stream-endpoint"

        def _call(
            self,
            messages: list[BaseMessage],
            stop: list[str] | None = None,
            run_manager: CallbackManagerForLLMRun | None = None,
            **kwargs: Any,
        ) -> str:
            return "Databricks"

        def _stream(
            self,
            messages: list[BaseMessage],
            stop: list[str] | None = None,
            run_manager: CallbackManagerForLLMRun | None = None,
            **kwargs: Any,
        ) -> Iterator[ChatGenerationChunk]:
            for chunk_content, finish_reason in [
                ("Da", None),
                ("tab", None),
                ("ricks", "stop"),
            ]:
                chunk = ChatGenerationChunk(
                    message=AIMessageChunk(content=chunk_content),
                    generation_info={"finish_reason": finish_reason},
                )
                if run_manager:
                    run_manager.on_llm_new_token(chunk.text, chunk=chunk)

                yield chunk

        @property
        def _llm_type(self) -> str:
            return "fake chat model"

    return FakeChatStreamModel(endpoint_name="fake-stream-endpoint")


@skip_if_v1
@pytest.mark.parametrize("provide_signature", [True, False])
def test_simple_chat_model_stream_inference(fake_chat_stream_model, provide_signature):
    # TODO: Migrate to models-from-code
    input_example = {
        "messages": [
            {"role": "system", "content": "You are a helpful assistant."},
            {"role": "assistant", "content": "What would you like to ask?"},
            {"role": "user", "content": "Who owns MLflow?"},
        ]
    }
    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            fake_chat_stream_model,
            name="model",
        )

    if provide_signature:
        signature = infer_signature(model_input=input_example)
        with mlflow.start_run():
            model_with_siginature_info = mlflow.langchain.log_model(
                fake_chat_stream_model, name="model", signature=signature
            )
    else:
        with mlflow.start_run():
            model_with_siginature_info = mlflow.langchain.log_model(
                fake_chat_stream_model, name="model", input_example=input_example
            )

    for model_uri in [model_info.model_uri, model_with_siginature_info.model_uri]:
        loaded_model = mlflow.pyfunc.load_model(model_uri)

        chunk_iter = loaded_model.predict_stream(input_example)

        finish_reason = "stop"

        with mock.patch("time.time", return_value=1677858242):
            chunks = list(chunk_iter)

            for chunk in chunks:
                assert "id" in chunk, "chunk id is lost."
                chunk["id"] = None

            assert chunks == [
                {
                    "id": None,
                    "object": "chat.completion.chunk",
                    "created": 1677858242,
                    "model": "",
                    "choices": [
                        {
                            "index": 0,
                            "finish_reason": None,
                            "delta": {"role": "assistant", "content": "Da"},
                        }
                    ],
                },
                {
                    "id": None,
                    "object": "chat.completion.chunk",
                    "created": 1677858242,
                    "model": "",
                    "choices": [
                        {
                            "index": 0,
                            "finish_reason": None,
                            "delta": {"role": "assistant", "content": "tab"},
                        }
                    ],
                },
                {
                    "id": None,
                    "object": "chat.completion.chunk",
                    "created": 1677858242,
                    "model": "",
                    "choices": [
                        {
                            "index": 0,
                            "finish_reason": finish_reason,
                            "delta": {"role": "assistant", "content": "ricks"},
                        }
                    ],
                },
            ]


@skip_if_v1
def test_simple_chat_model_stream_with_callbacks(fake_chat_stream_model):
    # TODO: Migrate to models-from-code
    class TestCallbackHandler(BaseCallbackHandler):
        def __init__(self):
            super().__init__()
            self.num_llm_start_calls = 0

        def on_llm_start(
            self,
            serialized: dict[str, Any],
            prompts: list[str],
            **kwargs: Any,
        ) -> Any:
            self.num_llm_start_calls += 1

    prompt = ChatPromptTemplate.from_template("What's your favorite {industry} company?")
    chain = prompt | fake_chat_stream_model | StrOutputParser()
    # Test the basic functionality of the chain
    assert chain.invoke({"industry": "tech"}) == "Databricks"

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            chain, name="model_path", input_example={"industry": "tech"}
        )

    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)

    callback_handler1 = TestCallbackHandler()
    callback_handler2 = TestCallbackHandler()

    # Ensure handlers have not been called yet
    assert callback_handler1.num_llm_start_calls == 0
    assert callback_handler2.num_llm_start_calls == 0

    stream = pyfunc_loaded_model._model_impl._predict_stream_with_callbacks(
        {"industry": "tech"},
        callback_handlers=[callback_handler1, callback_handler2],
    )
    assert list(stream) == ["Da", "tab", "ricks"]

    # Test that the callback handlers were called
    assert callback_handler1.num_llm_start_calls == 1
    assert callback_handler2.num_llm_start_calls == 1


@skip_if_v1
def test_langchain_model_save_load_with_listeners(fake_chat_model):
    # Migrate this to models-from-code
    prompt = ChatPromptTemplate.from_messages([
        ("system", "You are a helpful assistant."),
        MessagesPlaceholder(variable_name="history"),
        ("human", "{question}"),
    ])

    def retrieve_history(input):
        return {"history": [], "question": input["question"], "name": input["name"]}

    chain = (
        {"question": itemgetter("question"), "name": itemgetter("name")}
        | (RunnableLambda(retrieve_history) | prompt | fake_chat_model).with_listeners()
        | StrOutputParser()
        | RunnablePassthrough()
    )
    input_example = {"question": "Who owns MLflow?", "name": ""}
    assert chain.invoke(input_example) == "Databricks"

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            chain, name="model_path", input_example=input_example
        )
    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    assert loaded_model.invoke(input_example) == "Databricks"
    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    assert pyfunc_loaded_model.predict(input_example) == ["Databricks"]

    inference_payload = load_serving_example(model_info.model_uri)
    response = pyfunc_serve_and_score_model(
        model_info.model_uri,
        data=inference_payload,
        content_type=pyfunc_scoring_server.CONTENT_TYPE_JSON,
        extra_args=["--env-manager", "local"],
    )
    assert PredictionsResponse.from_json(response.content.decode("utf-8")) == {
        "predictions": ["Databricks"]
    }


@pytest.mark.parametrize("env_var", ["MLFLOW_ENABLE_TRACE_IN_SERVING", "ENABLE_MLFLOW_TRACING"])
def test_langchain_model_not_inject_callback_when_disabled(monkeypatch, model_path, env_var):
    # Emulate the model serving environment
    monkeypatch.setenv("IS_IN_DB_MODEL_SERVING_ENV", "true")

    # Disable tracing
    monkeypatch.setenv(env_var, "false")

    mlflow.langchain.save_model(SIMPLE_MODEL_CODE_PATH, model_path)

    loaded_model = mlflow.pyfunc.load_model(model_path)
    loaded_model.predict({"product": "shoe"})

    # Trace should be logged to the inference table
    from mlflow.tracing.export.inference_table import _TRACE_BUFFER

    assert _TRACE_BUFFER == {}


@pytest.mark.parametrize(
    "chain_path",
    [
        os.path.abspath("tests/langchain/sample_code/no_config/chain.py"),
        "tests/langchain/../langchain/sample_code/no_config/chain.py",
    ],
)
def test_save_model_as_code_correct_streamable(chain_path):
    input_example = {"messages": [{"role": "user", "content": "Who owns MLflow?"}]}
    answer = "Databricks"
    artifact_path = "model_path"
    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            chain_path,
            name=artifact_path,
            input_example=input_example,
        )

    assert model_info.flavors["langchain"]["streamable"] is True
    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)

    with mock.patch("time.time", return_value=1677858242):
        assert pyfunc_loaded_model._model_impl._predict_with_callbacks(input_example) == {
            "id": None,
            "object": "chat.completion",
            "created": 1677858242,
            "model": "",
            "choices": [
                {
                    "index": 0,
                    "message": {
                        "role": "assistant",
                        "content": "Databricks",
                    },
                    "finish_reason": None,
                }
            ],
            "usage": {
                "prompt_tokens": None,
                "completion_tokens": None,
                "total_tokens": None,
            },
        }

    inference_payload = load_serving_example(model_info.model_uri)
    response = pyfunc_serve_and_score_model(
        model_info.model_uri,
        data=inference_payload,
        content_type=pyfunc_scoring_server.CONTENT_TYPE_JSON,
        extra_args=["--env-manager", "local"],
    )
    # avoid minor diff of created time in the response
    prediction_result = json.loads(response.content.decode("utf-8"))
    prediction_result[0]["created"] = 123
    expected_prediction = try_transform_response_to_chat_format(answer)
    expected_prediction["created"] = 123
    assert prediction_result == [expected_prediction]

    pyfunc_model_path = _download_artifact_from_uri(model_info.model_uri)
    reloaded_model = Model.load(os.path.join(pyfunc_model_path, "MLmodel"))
    assert reloaded_model.resources["databricks"] == {
        "serving_endpoint": [{"name": "fake-endpoint"}]
    }


@skip_if_v1
def test_save_load_langchain_binding(fake_chat_model):
    runnable_binding = RunnableBinding(bound=fake_chat_model, kwargs={"stop": ["-"]})
    model = runnable_binding | StrOutputParser()
    assert model.invoke("Say something") == "Databricks"

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            model, name="model_path", input_example="Say something"
        )
    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    assert loaded_model.first.kwargs == {"stop": ["-"]}
    assert loaded_model.invoke("hello") == "Databricks"
    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    assert pyfunc_loaded_model.predict("hello") == ["Databricks"]

    inference_payload = load_serving_example(model_info.model_uri)
    response = pyfunc_serve_and_score_model(
        model_info.model_uri,
        data=inference_payload,
        content_type=pyfunc_scoring_server.CONTENT_TYPE_JSON,
        extra_args=["--env-manager", "local"],
    )
    assert PredictionsResponse.from_json(response.content.decode("utf-8")) == {
        "predictions": ["Databricks"]
    }


@skip_if_v1
def test_save_load_langchain_binding_llm_with_tool():
    from langchain_core.tools import tool

    # We need to use ChatOpenAI from langchain_openai as community one does not support bind_tools
    from langchain_openai import ChatOpenAI

    @tool
    def add(a: int, b: int) -> int:
        """Adds a and b.

        Args:
            a: first int
            b: second int
        """
        return a + b

    runnable_binding = ChatOpenAI(temperature=0.9).bind_tools([add])
    model = runnable_binding | StrOutputParser()
    expected_output = '[{"role": "user", "content": "hello"}]'
    assert model.invoke("hello") == expected_output

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(model, name="model_path", input_example="hello")

    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    assert loaded_model.invoke("hello") == expected_output
    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    assert pyfunc_loaded_model.predict("hello") == [expected_output]


@skip_if_v1
def test_langchain_bindings_save_load_with_config_and_types(fake_chat_model):
    class CustomCallbackHandler(BaseCallbackHandler):
        def __init__(self):
            self.count = 0

        def on_chain_start(
            self, serialized: dict[str, Any], inputs: dict[str, Any], **kwargs: Any
        ) -> None:
            self.count += 1

        def on_chain_end(self, outputs: dict[str, Any], **kwargs: Any) -> None:
            self.count += 1

    model = fake_chat_model | StrOutputParser()
    callback = CustomCallbackHandler()
    model = model.with_config(run_name="test_run", callbacks=[callback]).with_types(
        input_type=str, output_type=str
    )
    assert model.invoke("Say something") == "Databricks"
    assert callback.count == 4

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(model, name="model_path", input_example="hello")
    loaded_model = mlflow.langchain.load_model(model_info.model_uri)
    assert loaded_model.config["run_name"] == "test_run"
    assert loaded_model.custom_input_type == str
    assert loaded_model.custom_output_type == str
    callback = loaded_model.config["callbacks"][0]
    assert loaded_model.invoke("hello") == "Databricks"
    assert callback.count > 8  # accumulated count (inside model logging we also call the callbacks)
    pyfunc_loaded_model = mlflow.pyfunc.load_model(model_info.model_uri)
    assert pyfunc_loaded_model.predict("hello") == ["Databricks"]

    inference_payload = load_serving_example(model_info.model_uri)
    response = pyfunc_serve_and_score_model(
        model_info.model_uri,
        data=inference_payload,
        content_type=pyfunc_scoring_server.CONTENT_TYPE_JSON,
        extra_args=["--env-manager", "local"],
    )
    assert PredictionsResponse.from_json(response.content.decode("utf-8")) == {
        "predictions": ["Databricks"]
    }


@pytest.mark.parametrize(
    "chain_path",
    [
        os.path.abspath("tests/langchain/sample_code/chain.py"),
        "tests/langchain/../langchain/sample_code/chain.py",
    ],
)
@pytest.mark.parametrize(
    "model_config",
    [
        os.path.abspath("tests/langchain/sample_code/config.yml"),
        "tests/langchain/../langchain/sample_code/config.yml",
    ],
)
def test_load_chain_with_model_config_overrides_saved_config(chain_path, model_config):
    input_example = {
        "messages": [
            {
                "role": "user",
                "content": "What is a good name for a company that makes MLflow?",
            }
        ]
    }
    artifact_path = "model_path"
    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            chain_path,
            name=artifact_path,
            input_example=input_example,
            model_config=model_config,
        )

    with mock.patch("mlflow.langchain.model._load_model_code_path") as load_model_code_path_mock:
        mlflow.pyfunc.load_model(model_info.model_uri, model_config={"embedding_size": 2})
        args, kwargs = load_model_code_path_mock.call_args
        assert args[1] == {
            "embedding_size": 2,
            "llm_prompt_template": "Answer the following question based on the "
            "context: {context}\nQuestion: {question}",
            "not_used_array": [
                1,
                2,
                3,
            ],
            "response": "Databricks",
        }


@skip_if_v1
@pytest.mark.parametrize("streamable", [True, False, None])
def test_langchain_model_streamable_param_in_log_model(streamable, fake_chat_model):
    # TODO: Migrate to models-from-code
    prompt = ChatPromptTemplate.from_template("What's your favorite {industry} company?")
    chain = prompt | fake_chat_model | StrOutputParser()

    runnable = RunnableParallel({"llm": lambda _: "completion"})

    for model in [chain, runnable]:
        with mock.patch("mlflow.langchain.model._save_model"), mlflow.start_run():
            model_info = mlflow.langchain.log_model(
                model,
                name="model",
                streamable=streamable,
                pip_requirements=[],
            )

            expected = (streamable is None) or streamable
            assert model_info.flavors["langchain"]["streamable"] is expected


@pytest.fixture
def model_type(request):
    return lc_runnables_types()[request.param]


@skip_if_v1
@pytest.mark.parametrize("streamable", [True, False, None])
@pytest.mark.parametrize("model_type", range(len(lc_runnables_types())), indirect=True)
def test_langchain_model_streamable_param_in_log_model_for_lc_runnable_types(
    streamable, model_type
):
    with mock.patch("mlflow.langchain.model._save_model"), mlflow.start_run():
        model = mock.MagicMock(spec=model_type)
        assert hasattr(model, "stream") is True
        model_info = mlflow.langchain.log_model(
            model,
            name="model",
            streamable=streamable,
            pip_requirements=[],
        )

        expected = (streamable is None) or streamable
        assert model_info.flavors["langchain"]["streamable"] is expected

        del model.stream
        assert hasattr(model, "stream") is False
        model_info = mlflow.langchain.log_model(
            model,
            name="model",
            streamable=streamable,
            pip_requirements=[],
        )
        assert model_info.flavors["langchain"]["streamable"] is bool(streamable)


@skip_if_v1
def test_agent_executor_model_with_messages_input():
    question = {"messages": [{"role": "user", "content": "Who owns MLflow?"}]}

    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            os.path.abspath("tests/langchain/agent_executor/chain.py"),
            name="model_path",
            input_example=question,
            model_config=os.path.abspath("tests/langchain/agent_executor/config.yml"),
        )
    native_model = mlflow.langchain.load_model(model_info.model_uri)
    assert native_model.invoke(question)["output"] == "Databricks"
    pyfunc_model = mlflow.pyfunc.load_model(model_info.model_uri)
    # TODO: in the future we should fix this and output shouldn't be wrapped
    # The result is wrapped in a list because during signature enforcement we convert
    # input data to pandas dataframe, then inside _convert_llm_input_data
    # we convert pandas dataframe back to records, and a single row will be
    # wrapped inside a list.
    assert pyfunc_model.predict(question) == ["Databricks"]

    # Test stream output
    response = pyfunc_model.predict_stream(question)
    assert inspect.isgenerator(response)

    expected_response = [
        {
            "output": "Databricks",
            "messages": [
                {
                    "additional_kwargs": {},
                    "content": "Databricks",
                    "example": False,
                    "id": None,
                    "invalid_tool_calls": [],
                    "name": None,
                    "response_metadata": {},
                    "tool_calls": [],
                    "type": "ai",
                    "usage_metadata": None,
                }
            ],
        }
    ]
    assert list(response) == expected_response


def test_invoking_model_with_params():
    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            os.path.abspath("tests/langchain/sample_code/model_with_config.py"),
            name="model",
        )
    pyfunc_model = mlflow.pyfunc.load_model(model_info.model_uri)
    data = {"x": 0}
    pyfunc_model.predict(data)
    params = {"config": {"temperature": 3.0}}
    with mock.patch("mlflow.pyfunc._validate_prediction_input", return_value=(data, params)):
        # This proves the temperature is passed to the model
        with pytest.raises(MlflowException, match=r"Input should be less than or equal to 2"):
            pyfunc_model.predict(data=data, params=params)


def test_custom_resources(tmp_path):
    input_example = {
        "messages": [
            {
                "role": "user",
                "content": "What is a good name for a company that makes MLflow?",
            }
        ]
    }
    expected_resources = {
        "api_version": "1",
        "databricks": {
            "serving_endpoint": [
                {"name": "databricks-mixtral-8x7b-instruct"},
                {"name": "databricks-bge-large-en"},
                {"name": "azure-eastus-model-serving-2_vs_endpoint"},
            ],
            "vector_search_index": [{"name": "rag.studio_bugbash.databricks_docs_index"}],
            "sql_warehouse": [{"name": "testid"}],
            "function": [
                {"name": "rag.studio.test_function_a"},
                {"name": "rag.studio.test_function_b"},
            ],
        },
    }
    artifact_path = "model_path"
    chain_path = "tests/langchain/sample_code/chain.py"
    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            chain_path,
            name=artifact_path,
            input_example=input_example,
            model_config="tests/langchain/sample_code/config.yml",
            resources=[
                DatabricksServingEndpoint(endpoint_name="databricks-mixtral-8x7b-instruct"),
                DatabricksServingEndpoint(endpoint_name="databricks-bge-large-en"),
                DatabricksServingEndpoint(endpoint_name="azure-eastus-model-serving-2_vs_endpoint"),
                DatabricksVectorSearchIndex(index_name="rag.studio_bugbash.databricks_docs_index"),
                DatabricksSQLWarehouse(warehouse_id="testid"),
                DatabricksFunction(function_name="rag.studio.test_function_a"),
                DatabricksFunction(function_name="rag.studio.test_function_b"),
            ],
        )

        model_path = _download_artifact_from_uri(model_info.model_uri)
        reloaded_model = Model.load(os.path.join(model_path, "MLmodel"))
        assert reloaded_model.resources == expected_resources

    yaml_file = tmp_path.joinpath("resources.yaml")
    with open(yaml_file, "w") as f:
        f.write(
            """
            api_version: "1"
            databricks:
                vector_search_index:
                - name: rag.studio_bugbash.databricks_docs_index
                serving_endpoint:
                - name: databricks-mixtral-8x7b-instruct
                - name: databricks-bge-large-en
                - name: azure-eastus-model-serving-2_vs_endpoint
                sql_warehouse:
                - name: testid
                function:
                - name: rag.studio.test_function_a
                - name: rag.studio.test_function_b
            """
        )

    artifact_path_2 = "model_path_2"
    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            chain_path,
            name=artifact_path_2,
            input_example=input_example,
            model_config="tests/langchain/sample_code/config.yml",
            resources=yaml_file,
        )

        model_path = _download_artifact_from_uri(model_info.model_uri)
        reloaded_model = Model.load(os.path.join(model_path, "MLmodel"))
        assert reloaded_model.resources == expected_resources


def test_pyfunc_converts_chat_request_for_non_chat_model():
    input_example = {"messages": [{"role": "user", "content": "Hello"}]}
    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            lc_model=SIMPLE_MODEL_CODE_PATH,
            input_example=input_example,
        )

    pyfunc_model = mlflow.pyfunc.load_model(model_info.model_uri)
    result = pyfunc_model.predict(input_example)
    # output are converted to chatResponse format
    assert isinstance(result[0]["choices"][0]["message"]["content"], str)

    response = pyfunc_model.predict_stream(input_example)
    assert inspect.isgenerator(response)
    assert isinstance(list(response)[0]["choices"][0]["delta"]["content"], str)


@skip_if_v1
def test_pyfunc_should_not_convert_chat_request_if_env_var_is_set_to_false(monkeypatch):
    monkeypatch.setenv(MLFLOW_CONVERT_MESSAGES_DICT_FOR_LANGCHAIN.name, "false")

    # This model is an example when the model expects a chat request
    # format input, but the input should not be converted to List[BaseMessage]
    model = RunnablePassthrough.assign(problem=lambda x: x["messages"][-1]["content"]) | itemgetter(
        "problem"
    )
    input_example = {"messages": [{"role": "user", "content": "Databricks"}]}
    assert model.invoke(input_example) == "Databricks"

    # pyfunc model can accepts chat request format even the chain
    # itself does not accept it, but we need to use the correct
    # input example to infer model signature
    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(model, input_example=input_example)

    pyfunc_model = mlflow.pyfunc.load_model(model_info.model_uri)
    result = pyfunc_model.predict(input_example)
    assert result == ["Databricks"]

    # Test stream output
    response = pyfunc_model.predict_stream(input_example)
    assert inspect.isgenerator(response)
    assert list(response) == ["Databricks"], list(response)


def test_log_langchain_model_with_prompt():
    mlflow.register_prompt(
        name="qa_prompt",
        template="What is a good name for a company that makes {{product}}?",
        commit_message="Prompt for generating company names",
    )
    mlflow.set_prompt_alias("qa_prompt", alias="production", version=1)

    mlflow.register_prompt(name="another_prompt", template="Hi")

    # If the model code involves `mlflow.load_prompt()` call, the prompt version
    # should be automatically logged to the Run
    with mlflow.start_run():
        model_info = mlflow.langchain.log_model(
            os.path.abspath("tests/langchain/sample_code/chain_with_mlflow_prompt.py"),
            name="model",
            # Manually associate another prompt
            prompts=["prompts:/another_prompt/1"],
        )

    # Check that prompts were linked to the run via the linkedPrompts tag
    from mlflow.tracing.constant import TraceTagKey

    run = mlflow.MlflowClient().get_run(model_info.run_id)
    linked_prompts_tag = run.data.tags.get(TraceTagKey.LINKED_PROMPTS)
    assert linked_prompts_tag is not None

    linked_prompts = json.loads(linked_prompts_tag)
    assert len(linked_prompts) == 2
    assert {p["name"] for p in linked_prompts} == {"qa_prompt", "another_prompt"}

    prompt = mlflow.load_prompt("qa_prompt", 1)
    assert prompt.aliases == ["production"]

    prompt = mlflow.load_prompt("another_prompt", 1)

    pyfunc_model = mlflow.pyfunc.load_model(model_info.model_uri)
    response = pyfunc_model.predict({"product": "shoe"})
    # Fake OpenAI server echo the input
    assert (
        response
        == '[{"role": "user", "content": "What is a good name for a company that makes shoe?"}]'
    )


def test_predict_with_callbacks_with_tracing(monkeypatch):
    # Simulate the model serving environment
    monkeypatch.setenv("IS_IN_DB_MODEL_SERVING_ENV", "true")
    monkeypatch.setenv("ENABLE_MLFLOW_TRACING", "true")
    mlflow.tracing.reset()

    model_info = mlflow.langchain.log_model(
        os.path.abspath("tests/langchain/sample_code/workflow.py"),
        name="model_path",
        input_example={"messages": [{"role": "user", "content": "What is MLflow?"}]},
    )
    # serving environment only reads from this environment variable
    monkeypatch.setenv("MLFLOW_EXPERIMENT_ID", mlflow.last_logged_model().experiment_id)

    pyfunc_model = mlflow.pyfunc.load_model(model_info.model_uri)

    request_id = "mock_request_id"
    tracer = MlflowLangchainTracer(prediction_context=Context(request_id))
    input_example = {"messages": [{"role": "user", "content": TEST_CONTENT}]}

    with mock.patch("mlflow.tracing.client.TracingClient.start_trace") as mock_start_trace:
        pyfunc_model._model_impl._predict_with_callbacks(
            data=input_example, callback_handlers=[tracer]
        )
        mlflow.flush_trace_async_logging()
        mock_start_trace.assert_called_once()
        trace_info = mock_start_trace.call_args[0][0]
        assert trace_info.client_request_id == request_id
        assert trace_info.request_metadata[TraceMetadataKey.MODEL_ID] == model_info.model_id


@pytest.mark.skipif(not IS_LANGCHAIN_v1, reason="The test is only for langchain>=1 versions")
def test_langchain_v1_save_model_as_pickle_error():
    model = create_openai_runnable()
    with mlflow.start_run():
        with pytest.raises(
            MlflowException,
            match="LangChain v1 onward only supports models-from-code",
        ):
            mlflow.langchain.log_model(
                model, name="langchain_model", input_example={"product": "MLflow"}
            )