🚀open-souce LLMs benchmark

Evaluate the capabilities of open-source LLMs in agent, tool calling, formatted output, long context retrieval, multilingual support, coding, mathematics, and custom tasks.

Example

🤖Agent Task

The ReAct Agent can access 5 functions. There are 10 questions to be solved, 4 of which are simple questions that can be solved using a single function, and 6 of which are complicated questions that require the agent to use multiple steps to solve.

The score ranges from 1 to 5, with 5 representing complete correctness. Here is an screen shot while running evaluation.

🧐Retrieval Task

Insert the needle(answer) into a haystack(long context) and ask the model retrieval the question based on the long context.

🗣️Format output Task

Evaluate the model's ability to repond in specified format, such as JSON, Number, Python, etc.

BenchMark Evaluation

Supported:

• 🤖Agent, evaluate whether the model can accurately select tools or functions for invocation and follow the ReAct pattern to solve problems.
• 🗣️Formated output, evaluate whether the model can output content in required formats such as JSON, Single Number, Code Bloch, etc.
• 🧐Long context retrieval, capability to retrieval correct fact from a long context.

Plan:

• 🇺🇸🇨🇳Multilingual, capability to understand and respond in different languages.
• ⌨️coding, capability to solve complicated promblem with code.
• ∞Mathematics, capability to solve mathematic problem w/ or w/o code interpreter
• 😀Custom Task, easily define and evaluate any specific task which you concern.

Install

Install from pypi:

pip install open_llm_benchmark

Install from github repo:

git clone [email protected]:EvilPsyCHo/Open-LLM-Benchmark.git
cd Open-LLM-Benchmark
python setup.py install

Supoorted Backend

• Huggingface transformers
• llama-cpp-pyton
• vLLM
• OpenAI

Contribute

Feel free to contribute this project!

• more backend such as Anthropic, ollama, etc.
• more tasks.
• more evaluation data.
• visualize the evaluation result.
• etc.