This folder contains code and resources to run experiments and evaluations.
To better organize the evaluation folder, we should follow the rules below:
- Each subfolder contains a specific benchmark or experiment. For example,
evaluation/swe_benchshould contain all the preprocessing/evaluation/analysis scripts. - Raw data and experimental records should not be stored within this repo.
- For model outputs, they should be stored at this huggingface space for visualization.
- Important data files of manageable size and analysis scripts (e.g., jupyter notebooks) can be directly uploaded to this repo.
To learn more about how to integrate your benchmark into OpenHands, check out tutorial here.
- SWE-Bench:
evaluation/swe_bench - HumanEvalFix:
evaluation/humanevalfix - BIRD:
evaluation/bird - BioCoder:
evaluation/ml_bench - ML-Bench:
evaluation/ml_bench - APIBench:
evaluation/gorilla - ToolQA:
evaluation/toolqa
- WebArena:
evaluation/webarena - MiniWob++:
evaluation/miniwob
- GAIA:
evaluation/gaia - GPQA:
evaluation/gpqa - AgentBench:
evaluation/agent_bench - MINT:
evaluation/mint - Entity deduction Arena (EDA):
evaluation/EDA - ProofWriter:
evaluation/logic_reasoning
Please follow instruction here to setup your local development environment and LLM.
OpenHands in development mode uses config.toml to keep track of most configurations.
Here's an example configuration file you can use to define and use multiple LLMs:
[llm]
# IMPORTANT: add your API key here, and set the model to the one you want to evaluate
model = "gpt-4o-2024-05-13"
api_key = "sk-XXX"
[llm.eval_gpt4_1106_preview_llm]
model = "gpt-4-1106-preview"
api_key = "XXX"
temperature = 0.0
[llm.eval_some_openai_compatible_model_llm]
model = "openai/MODEL_NAME"
base_url = "https://OPENAI_COMPATIBLE_URL/v1"
api_key = "XXX"
temperature = 0.0Check this huggingface space for visualization of existing experimental results.
You can start your own fork of our huggingface evaluation outputs and submit a PR of your evaluation results to our hosted huggingface repo via PR following the guide here.