flow-eval

Technical Report | Model Weights | HuggingFace Space | Evaluation Code | Tutorials

flow-eval is an evaluation library for LLM applications that supports LLM-as-a-judge, function-based, and similarity-based evaluation. It provides flexible evals, multiple inference backends, and seamless integration with popular frameworks.

Installation

Install flow-eval using pip:

pip install -e ".[vllm,hf]"
pip install 'flash_attn>=2.6.3' --no-build-isolation

Extras available:

• dev to install development dependencies
• hf to install Hugging Face Transformers dependencies
• vllm to install vLLM dependencies
• llamafile to install Llamafile dependencies
• baseten to install Baseten dependencies
• similarity to install similarity evaluation dependencies (torch, sentence-transformers)
• openai to install OpenAI API dependencies

Quick Start

Here's a simple example to get you started:

from flow_eval import LMEvaluator
from flow_eval.lm.models import Vllm, Llamafile, Hf
from flow_eval.core import EvalInput
from flow_eval.lm.metrics import RESPONSE_FAITHFULNESS_5POINT
from IPython.display import Markdown, display

# If you are running on an Ampere GPU or newer, create a model using VLLM
model = Vllm()

# If you have other applications open taking up VRAM, you can use less VRAM by setting gpu_memory_utilization to a lower value.
# model = Vllm(gpu_memory_utilization=0.70)

# Or if not running on Ampere GPU or newer, create a model using no flash attn and Hugging Face Transformers
# model = Hf(flash_attn=False)

# Or create a model using Llamafile if not running an Nvidia GPU & running a Silicon MacOS for example
# model = Llamafile()

# Initialize the evaluator
faithfulness_evaluator = LMEvaluator(
    eval=RESPONSE_FAITHFULNESS_5POINT,
    model=model
)

# Sample to evaluate
query = """..."""
context = """...""""
response = """..."""

# Create an EvalInput
# We want to evaluate the response to the customer issue based on the context and the user instructions
eval_input = EvalInput(
    inputs=[
        {"query": query},
        {"context": context},
    ],
    output={"response": response},
)

# Run the evaluation
result = faithfulness_evaluator.evaluate(eval_input, save_results=False)

# Display the result
display(Markdown(f"__Feedback:__\n{result.feedback}\n\n__Score:__\n{result.score}"))

Usage

Inference Options

The library supports multiple inference backends to accommodate different hardware configurations and performance needs:

1. vLLM:

   • Best for NVIDIA GPUs with Ampere architecture or newer (e.g., RTX 3000 series, A100, H100)
   • Offers the highest performance and throughput
   • Requires CUDA-compatible GPU

   from flow_eval.lm.models import Vllm
   
   model = Vllm()

2. Hugging Face Transformers:

   • Compatible with a wide range of hardware, including older NVIDIA GPUs
   • Supports CPU inference (slower but universally compatible)
   • It is slower than vLLM but generally compatible with more hardware.

   If you are running on an Ampere GPU or newer:

   from flow_eval.lm.models import Hf
   
   model = Hf()

   If you are not running on an Ampere GPU or newer, disable flash attention:

   from flow_eval.lm.models import Hf
   
   model = Hf(flash_attn=False)

3. Llamafile:

   • Ideal for non-NVIDIA hardware, including Apple Silicon
   • Provides good performance on CPUs
   • Self-contained, easy to deploy option

   from flow_eval.lm.models import Llamafile
   
   model = Llamafile()

4. Baseten:

   • Remote execution.
   • Machine independent.
   • Improved concurrency patterns for larger workloads.

from flow_eval.lm.models import Baseten

model = Baseten()

For detailed information on using Baseten, visit the Baseten readme.

5. OpenAI:
   • Uses OpenAI's API for remote evaluation
   • Supports various OpenAI models including GPT-4
   • Requires OPENAI_API_KEY environment variable

from flow_eval.lm.models import OpenAIModel
import os

# Set your API key
os.environ["OPENAI_API_KEY"] = "your-api-key-here"

model = OpenAIModel(model="gpt-4")

Install with: pip install flow-eval[openai]

Choose the inference backend that best matches your hardware and performance requirements. The library provides a unified interface for all these options, making it easy to switch between them as needed.

Evals

Flow-eval-v0.1 was trained to handle any custom metric that can be expressed as a combination of evaluation criteria and rubric, and required inputs and outputs.

Pre-defined Evals

For convenience, flow-eval library comes with pre-defined metrics such as RESPONSE_CORRECTNESS or RESPONSE_FAITHFULNESS. You can check the full list by running:

from flow_eval import list_all_lm_evals

list_all_lm_evals()

Batched Evaluations

For efficient processing of multiple inputs, you can use the batch_evaluate method:

# Read the sample data
import json
from flow_eval import LMEvaluator
from flow_eval.core import EvalInput
from flow_eval.lm.models import Vllm
from flow_eval.lm.metrics import RESPONSE_FAITHFULNESS_5POINT
from IPython.display import Markdown, display

# Initialize the model
model = Vllm()

# Initialize the evaluator
faithfulness_evaluator = LMEvaluator(
    eval=RESPONSE_FAITHFULNESS_5POINT,
    model=model
)

# Load some sampledata
with open("sample_data/csr_assistant.json", "r") as f:
    data = json.load(f)

# Create a list of inputs and outputs
inputs_batch = [
    [
        {"query": sample["query"]},
        {"context": sample["context"]},
    ]
    for sample in data
]
outputs_batch = [{"response": sample["response"]} for sample in data]

# Create a list of EvalInput
eval_inputs_batch = [EvalInput(inputs=inputs, output=output) for inputs, output in zip(inputs_batch, outputs_batch)]

# Run the batch evaluation
results = faithfulness_evaluator.batch_evaluate(eval_inputs_batch, save_results=False)

# Visualizing the results
for i, result in enumerate(results):
    display(Markdown(f"__Sample {i+1}:__"))
    display(Markdown(f"__Feedback:__\n{result.feedback}\n\n__Score:__\n{result.score}"))
    display(Markdown("---"))

Similarity-Based Evaluation

For evaluating semantic similarity between responses and expected outputs, you can use the AnswerSimilarityEvaluator:

from flow_eval import AnswerSimilarityEvaluator
from flow_eval.core import EvalInput

# Initialize the similarity evaluator
similarity_evaluator = AnswerSimilarityEvaluator(
    model_name="all-mpnet-base-v2",  # Sentence transformer model
    similarity_fn_name="cosine",      # cosine, dot, euclidean, or manhattan
    output_dir=None                   # Set to save results
)

# Create evaluation input with expected output
eval_input = EvalInput(
    inputs=[],
    output={"response": "The quick brown fox jumps over the lazy dog"},
    expected_output={"reference": "A fast brown fox leaps above a sleeping canine"}
)

# Run similarity evaluation
result = similarity_evaluator.evaluate(eval_input, save_results=False)
print(f"Similarity score: {result.score:.3f}")  # Range: 0.0 to 1.0

# Batch evaluation
eval_inputs = [
    EvalInput(
        inputs=[],
        output={"response": "Hello world"},
        expected_output={"reference": "Hi world"}
    ),
    EvalInput(
        inputs=[],
        output={"response": "Python programming"},
        expected_output={"reference": "Python coding"}
    )
]

results = similarity_evaluator.batch_evaluate(eval_inputs, save_results=False)
for i, result in enumerate(results):
    print(f"Sample {i+1} similarity: {result.score:.3f}")

The similarity evaluator requires the similarity extra:

pip install flow-eval[similarity]

Advanced Usage

Warning

There exists currently a reported issue with Phi-3 models that produces gibberish outputs with contexts longer than 4096 tokens, including input and output. This issue has been recently fixed in the transformers library so we recommend using the Hf() model configuration for longer contexts at the moment. For more details, refer to: #33129 and #6135

Custom Evals

Create your own evaluation metrics:

from flow_eval.lm import LMEval, RubricItem

custom_metric = LMEval(
    name="My Custom Metric",
    criteria="Evaluate based on X, Y, and Z.",
    rubric=[
        RubricItem(score=0, description="Poor performance"),
        RubricItem(score=1, description="Good performance"),
    ],
    input_columns=["query"],
    output_column="response"
)

evaluator = LMEvaluator(eval=custom_metric, config="Flow-eval-v0.1-AWQ")

Integrations

We support an integration with Llama Index evaluation module and Haystack:

• Llama Index tutorial
• Haystack tutorial

Note that we are currently working on adding more integrations with other frameworks in the near future.

Development Setup

1. Clone the repository:

   git clone https://github.com/flowaicom/flow-eval.git
   cd flow-eval

2. Create a virtual environment:

   virtualenv ./.venv

   or

   python -m venv ./.venv

3. Activate the virtual environment:

   • On Windows:

     venv\Scripts\activate

   • On macOS and Linux:

     source venv/bin/activate

4. Install the package in editable mode with development dependencies:

   pip install -e ".[dev]"

   or

   pip install -e ".[dev,vllm]"

   for vLLM support.

5. Set up pre-commit hooks:

   pre-commit install

6. Make sure you have trufflehog installed:

   # make trufflehog available in your path
   # macos
   brew install trufflehog
   # linux
   curl -sSfL https://raw.githubusercontent.com/trufflesecurity/trufflehog/main/scripts/install.sh | sh -s -- -b /usr/local/bin
   # nix
   nix profile install nixpkgs#trufflehog

7. Run pre-commit on all files:

   pre-commit run --all-files

8. You're now ready to start developing! You can run the main script with:

   python -m flow_eval

Remember to always activate your virtual environment when working on the project. To deactivate the virtual environment when you're done, simply run:

deactivate

Running Tests

To run the tests for Flow-eval, follow these steps:

1. Navigate to the root directory of the project in your terminal.

2. Run the tests using pytest:

   pytest tests/

   This will discover and run all the tests in the tests/ directory.

3. Run different test categories:

   pytest tests/unit/              # Unit tests only
   pytest tests/e2e-local/         # Local end-to-end tests
   pytest tests/e2e-cloud-gpu/     # Cloud GPU tests (requires GPU)

4. Run tests with coverage:

   pytest tests/unit --cov=./

5. For more verbose output, you can use the -v flag:

   pytest -v tests/

Contributing

Contributions to flow-eval are welcome! Please follow these steps:

1. Fork the repository
2. Create your feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add some AmazingFeature')
4. Push to the branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

Please ensure that your code adheres to the project's coding standards and passes all tests.

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Acknowledgments

flow-eval is developed and maintained by the Flow AI team. We appreciate the contributions and feedback from the AI community in making this tool more robust and versatile.