090c911a50
* change runtime to be synchronous * fix test runtime with the new interface * fix arg * fix eval * fix missing config attribute * fix plugins * fix on_event by revert it back to async * update upload_file endpoint * fix argument to upload file * remove unncessary async for eval; fix evaluation run in parallel * use asyncio to run controller for eval * revert file upload * truncate eval test result output
ToolQA Evaluation with OpenHands
This folder contains an evaluation harness we built on top of the original ToolQA (paper).
Setup Environment and LLM Configuration
Please follow instruction here to setup your local development environment and LLM.
Run Inference on ToolQA Instances
Make sure your Docker daemon is running, then run this bash script:
bash evaluation/toolqa/scripts/run_infer.sh [model_config] [git-version] [agent] [eval_limit] [dataset] [hardness] [wolfram_alpha_appid]
where model_config is mandatory, while all other arguments are optional.
model_config, e.g. llm, is the config group name for your
LLM settings, as defined in your config.toml.
git-version, e.g. HEAD, is the git commit hash of the OpenHands version you would
like to evaluate. It could also be a release tag like 0.6.2.
agent, e.g. CodeActAgent, is the name of the agent for benchmarks, defaulting
to CodeActAgent.
eval_limit, e.g. 10, limits the evaluation to the first eval_limit instances.
By default, the script evaluates 1 instance.
dataset, the dataset from ToolQA to evaluate from. You could choose from agenda, airbnb, coffee, dblp, flight, gsm8k, scirex, yelp for dataset. The default is coffee.
hardness, the hardness to evaluate. You could choose from easy and hard. The default is easy.
wolfram_alpha_appid is an optional argument. When given wolfram_alpha_appid, the agent will be able to access Wolfram Alpha's APIs.
Note: in order to use eval_limit, you must also set agent; in order to use dataset, you must also set eval_limit; in order to use hardness, you must also set dataset.
Let's say you'd like to run 10 instances using llm and CodeActAgent on coffee easy test,
then your command would be:
bash evaluation/toolqa/scripts/run_infer.sh llm CodeActAgent 10 coffee easy