libwebarena 0.0.5

This is an unofficial, use-at-your-own risks port of the webarena benchmark, for use as a standalone library package.

Warning: use at your own risks!

Unofficial WebArena port for compatibility with BrowserGym. Changes below.

More flexible/recent dependencies

• playwright~=1.39
• openai>=1
• transformers

Packaging into a single Python namespace

pip install libwebarena

import webarena
import webarena.browser_env
import webarena.agent
import webarena.evaluation_harness
import webarena.llms
import webarena.llms.providers

Making HTMLContentEvaluator idempotent (validate() should not alter the browser's state)