Skip to content

miniwob

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History
 
 

miniwob

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
scripts
scripts
 
 
README.md
README.md
 
 
__init__.py
__init__.py
 
 
get_avg_reward.py
get_avg_reward.py
 
 
run_infer.py
run_infer.py
 
 

README.md

WebArena Evaluation with OpenDevin Browsing Agents

This folder contains evaluation for MiniWoB++ benchmark, powered by BrowserGym for easy evaluation of how well an agent capable of browsing can perform on synthetic web browsing tasks.

Setup OpenDevin Environment

Please follow this document to setup local develop environment for OpenDevin.

Configure OpenDevin and your LLM

Create a config.toml file if it does not exist at the root of the workspace.

Add the following configurations:

[core]
max_iterations = 100
cache_dir = "/tmp/cache"
ssh_hostname = "localhost"

[sandbox]
box_type = "ssh"
timeout = 120

# TODO: Change these to the model you want to evaluate
[llm.eval_gpt4_1106_preview]
model = "gpt-4-1106-preview"
api_key = "XXX"
temperature = 0.0

[llm.eval_some_openai_compatible_model]
model = "openai/MODEL_NAME"
base_url = "https://OPENAI_COMPATIBLE_URL/v1"
api_key = "XXX"
temperature = 0.0

Setup MiniWoB++ Environment and Environment Variables of MiniWoB++

MiniWoB++ requires you to set up websites containing a static website that is accessible via URL to the machine running the OpenDevin agents.

  • Clone miniwob (use a specific frozen commit for reproducibility)
git clone [email protected]:Farama-Foundation/miniwob-plusplus.git
git -C "./miniwob-plusplus" reset --hard 7fd85d71a4b60325c6585396ec4f48377d049838
  • Setup Miniwob URL (change PATH_TO_MINIWOB_CLONED_REPO here to the absolute path to your miniwob-plusplus folder) in evaluation/miniwob/scripts/run_infer.sh
export MINIWOB_URL="file://<PATH_TO_MINIWOB_CLONED_REPO>/miniwob/html/miniwob/"

Test if your environment works

Access with browser the above MiniWoB URLs and see if they load correctly.

Run Evaluation

bash evaluation/miniwob/scripts/run_infer.sh

Results will be in evaluation/evaluation_outputs/outputs/miniwob/

To calculate the average reward, run:

poetry run python evaluation/miniwob/get_success_rate.py evaluation/evaluation_outputs/outputs/miniwob/SOME_AGENT/EXP_NAME/output.jsonl

Submit your evaluation results

You can start your own fork of our huggingface evaluation outputs and submit a PR of your evaluation results following the guide here.

BrowsingAgent V1.0 result

Tested on BrowsingAgent V1.0

MiniWoB++, 125 tasks (3 runs due to random init task), max step 10

  • GPT4o: 0.384, 0.416, 0.424, avg: 0.408
  • GPT3.5: 0.288, 0.256, 0.272, avg: 0.272