llama-index-packs-evaluator-benchmarker 0.3.0

llama-index packs evaluator_benchmarker integration

Evaluator Benchmarker Pack

A pack for quick computation of benchmark results of your own LLM evaluator on an Evaluation llama-dataset. Specifically, this pack supports benchmarking an appropriate evaluator on the following llama-datasets:

• LabelledEvaluatorDataset for single-grading evaluations
• LabelledPairwiseEvaluatorDataset for pairwise-grading evaluations

These llama-datasets can be downloaed from llama-hub.

CLI Usage

You can download llamapacks directly using llamaindex-cli, which comes installed with the llama-index python package:

llamaindex-cli download-llamapack EvaluatorBenchmarkerPack --download-dir ./evaluator_benchmarker_pack

You can then inspect the files at ./evaluator_benchmarker_pack and use them as a template for your own project!

Code Usage

You can download the pack to the ./evaluator_benchmarker_pack directory through python code as well. The sample script below demonstrates how to construct EvaluatorBenchmarkerPack using a LabelledPairwiseEvaluatorDataset downloaded from llama-hub and a PairwiseComparisonEvaluator that uses GPT-4 as the LLM. Note though that this pack can also be used on a LabelledEvaluatorDataset with a BaseEvaluator that performs single-grading evaluation — in this case, the usage flow remains the same.

from llama_index.core.llama_dataset import download_llama_dataset
from llama_index.core.llama_pack import download_llama_pack
from llama_index.core.evaluation import PairwiseComparisonEvaluator
from llama_index.llms.openai import OpenAI
from llama_index.core import ServiceContext

# download a LabelledRagDataset from llama-hub
pairwise_dataset = download_llama_dataset(
    "MiniMtBenchHumanJudgementDataset", "./data"
)

# define your evaluator
gpt_4_context = ServiceContext.from_defaults(
    llm=OpenAI(temperature=0, model="gpt-4"),
)
evaluator = PairwiseComparisonEvaluator(service_context=gpt_4_context)


# download and install dependencies
EvaluatorBenchmarkerPack = download_llama_pack(
    "EvaluatorBenchmarkerPack", "./pack"
)

# construction requires an evaluator and an eval_dataset
evaluator_benchmarker_pack = EvaluatorBenchmarkerPack(
    evaluator=evaluator,
    eval_dataset=pairwise_dataset,
    show_progress=True,
)

# PERFORM EVALUATION
benchmark_df = evaluator_benchmarker_pack.run()  # async arun() also supported
print(benchmark_df)

Output:

number_examples                1689
inconclusives                  140
ties                           379
agreement_rate_with_ties       0.657844
agreement_rate_without_ties    0.828205

Note that evaluator_benchmarker_pack.run() will also save the benchmark_df files in the same directory.

.
└── benchmark.csv