How can i setup langfuse enviroment for run experiment?

[noaft]

from langfuse import get_client
from loguru import logger
from dotenv import load_dotenv
load_dotenv()
# =========================
# ENVIRONMENT CONFIG
# =========================

Langfuse_client = get_client()

# =========================
# DUMMY TASK
# =========================
def task(*, item, **_):
    """A minimal task that just returns 1."""
    logger.info(f"Running task for item: {item}")
    return {"result": 1}

# =========================
# MAIN EXECUTION
# =========================
if __name__ == "__main__":
    dataset_name = "durian_pest_and_disease_image_retrieval_v0_0"
    logger.info(f"🚀 Starting minimal experiment: {dataset_name}")
    logger.info(f"🌍 Langfuse environment: {Langfuse_client._environment}")

    dataset = Langfuse_client.get_dataset(dataset_name)
    dataset.run_experiment(
        name="Dummy Experiment - Return 1",
        task=task,
        evaluators=[],
        metadata={"note": "This is a minimal test run"},
    )

    Langfuse_client.flush()
    logger.success("✅ Minimal experiment completed!")

I seted enviroments

LANGFUSE_SECRET_KEY=....
LANGFUSE_PUBLIC_KEY=....
LANGFUSE_HOST=....
LANGFUSE_BASE_URL=....
LANGFUSE_ENVIRONMENT=....

[noaft]

i setd LANGFUSE_TRACING_ENVIRONMENT=plant-doctor-v0_1_5
but it auto get default for this experiment

[jannikmaierhoefer]

Thanks @noaft, are you self-hosting Langfuse? If yes, which version are you running? We fixed an issue with the environment a couple of weeks ago and you might need to upgrade your langfuse instance.

[noaft]

okay, thanks for support @jannikmaierhoefer. I using langfuse verion 3.3.4 and self host langfuse with image docker langfuse/langfuse:3.

[jannikmaierhoefer]

Thanks. Please upgrade your Langfuse instance to fix the reported issue.

[noaft]

tks for support, but I have another way for people using this version.

for i, item in enumerate(dataset.items, 1):
    # Use item.run() to create a dedicated trace for this specific row.
    # Each iteration of this loop will create a new run in Langfuse.
    with item.run(
        run_name=f"Image retrieval {__version__} - {model_name} - {dt}",
        run_description="Image retrieval with Tulip + Milvus",
        metadata={"model": model_name, "version": __version__}
    ) as root_span:

        # 1. Execute your core logic by calling the task function.
        #    The `task` function is treated as a black box here.
        result = task(
            item.input,
            # base_url="...",
            # model_name=model_name,
            # limit=10
        )

        # 2. Update the root span with the inputs, outputs, and performance metrics.
        #    The `result` dictionary from your task should contain this data.
        root_span.update(
            input=item.input,
            output=result, # The full output from your task
            tags=[__version__, "production-test"],
            metadata={
                "duration": result.get("duration"),
                "s3_time": result.get("s3_time"),
                "embedding_time": result.get("embedding_time"),
                "milvus_time": result.get("milvus_time"),
            },
        )

we can using item run like a row evaluate, we can set everything like tag, ......

[jannikmaierhoefer]

Thanks for posting the workaround here @noaft !