llm-evaluation-in-reasoning 1.4.1

A project for evaluating reasoning capabilities in large language models (LLMs).

llm_evaluation_in_reasoning

A project for evaluating reasoning capabilities in large language models (LLMs).

Read this in other languages: English, 中文.

Run the eval

Install the package

pip install llm_evaluation_in_reasoning

Create the .env file

Create a .env file with the following:

OPENAI_API_KEY=<your key>
ANTHROPIC_API_KEY=<your key>
...

The api key you provided will be used to fetch the valid models supported by Litellm.

Run Instructions

Support GSM-Symbolic, GSM8K, MMLU, SimpleBench To run a benchmark:

llm_eval --model_name=ollama/qwen2.5:0.5b --dataset=SimpleBench # run llm_eval --help to see help information

Model support

Model support is based on Litellm, see the docs here Litellm Providers

Build the project

Setup Instructions

Clone the github repo and cd into it.

git clone https://github.com/ashengstd/llm_evaluation_in_reasoning.git
cd llm_evaluation_in_reasoning

Install uv:

The best way to install dependencies is to use uv. If you don't have it installed in your environment, you can install it with the following:

curl -LsSf https://astral.sh/uv/install.sh | sh # macOS and Linux
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex" # Windows

Sync the dependencies

uv sync --all-extra