Athina Client SDK

0.2.11 · active · verified Fri Apr 17

Athina Client is a lightweight Python SDK designed to interact with Athina's datasets and API. It enables users to log LLM inferences, evaluations, and experiments for monitoring, analysis, and debugging AI applications. The current version is 0.2.11, with new releases occurring regularly, typically a few times a month, especially for minor fixes and feature additions.

Common errors

```
requests.exceptions.HTTPError: 401 Client Error: Unauthorized for url:
```
cause The provided Athina API key is either missing, invalid, or expired.

fix Verify that the `ATHINA_API_KEY` environment variable is correctly set with a valid key, or ensure the `api_key` argument is passed correctly during `AthinaClient` initialization.
```
AttributeError: 'module' object has no attribute 'log_inference'
```
cause Attempting to call `log_inference` (or `log_evaluation`) directly from the `athina` module, which was deprecated in favor of calling it on an `AthinaClient` instance.

fix Initialize `athina_client = AthinaClient(...)` and then use `athina_client.log_inference(...)`.
```
TypeError: 'NoneType' object is not callable
```
cause This often occurs if the `AthinaClient` object was not correctly instantiated (e.g., due to an error during `__init__`) or if a method was called on a `None` object, which can happen if `AthinaClient(...)` failed to return an instance.

fix Check for errors during `AthinaClient` initialization, especially related to the `api_key`. Ensure the client object is valid before calling its methods.

Warnings

breaking The methods `log_inference` and `log_evaluation` were moved from directly under the `athina` module to be methods of the `AthinaClient` instance. If you were calling `athina.log_inference(...)`, you now need an instantiated client: `athina_client.log_inference(...)`.
Fix: Initialize `athina_client = AthinaClient(...)` and call methods like `athina_client.log_inference(...)`.
gotcha Failure to set the `ATHINA_API_KEY` environment variable or pass it directly to `AthinaClient(api_key=...)` will result in authentication errors (401 Unauthorized) when making API calls.
Fix: Set `export ATHINA_API_KEY='your-api-key'` in your environment or explicitly pass it during client initialization: `AthinaClient(api_key='your-api-key')`.
breaking The `log_experiment` method in `AthinaClient` removed the `batch_size` parameter.
Fix: Remove the `batch_size` argument if you are calling `log_experiment`.

Install

pip install athina-client Install latest version

Imports

AthinaClient
```
from athina.client import AthinaClient
```
AthinaInference
wrong:
```
from athina.client import AthinaInference
```
correct:
```
from athina.interfaces.athina import AthinaInference
```
Pydantic models like AthinaInference are typically found in the interfaces submodule, not directly under client.

Quickstart

This quickstart demonstrates how to initialize the Athina client and use its core functionality to log an LLM inference and a corresponding evaluation. Ensure the ATHINA_API_KEY environment variable is set.

import os
from athina.client import AthinaClient

# Initialize the Athina client with your API key
athina_client = AthinaClient(api_key=os.environ.get("ATHINA_API_KEY", ""))

# Example 1: Log an inference
try:
    response = athina_client.log_inference(
        prompt="What is the capital of France?",
        response="Paris",
        external_reference_id="my-app-inference-123",
        user_query="Capital question"
    )
    print(f"Inference logged successfully: {response.json()}")
except Exception as e:
    print(f"Error logging inference: {e}")

# Example 2: Log a simple evaluation
try:
    response = athina_client.log_evaluation(
        external_reference_id="my-app-evaluation-456",
        metric="correctness",
        score=1, # 1 for correct, 0 for incorrect
        details="The response was accurate.",
        language_model_id="gpt-4"
    )
    print(f"Evaluation logged successfully: {response.json()}")
except Exception as e:
    print(f"Error logging evaluation: {e}")

view raw JSON →