BiasGenerator
winner takes all survival simulation inspired by squid game glass bridge where the agents must negotiate,trade, collaborate and compete with one another for their own survival
Core trick
Runs a multi-round Glass Bridge tournament with scripted baseline policies and config-driven evaluation scripts.
Tags
Description
winner takes all survival simulation inspired by squid game glass bridge where the agents must negotiate,trade, collaborate and compete with one another for their own survival
README excerpt
# BiasGenerator This repo is focused on Glass Bridge simulation and tournament evaluation. Included: - a single-round `GlassBridgeEnv` - a multi-round `GlassBridgeTournamentEnv` - an OpenEnv-style HTTP server/client wrapper for the tournament game - scripted baseline policies - config-driven local and remote evaluation scripts - Dockerfiles for CPU and GPU-backed runs - artifact and log output under `artifacts/` ## Local setup Create a Python 3.11 virtual environment and install the local dependencies: ```bash python3.11 -m venv .venv source .venv/bin/activate python -m pip install --upgrade pip setuptools wheel pip install -r requirements/local.txt pip install -e . ``` ## Local runs Single-round Glass Bridge evaluation: ```bash python scripts/eval_glass_bridge.py --config configs/glass_bridge_local.yaml ``` Tournament evaluation: ```bash python scripts/eval_glass_bridge_tournament.py --config configs/glass_bridge_tournament_local.yaml ``` The tournament configs now use the OpenEnv-style transport by default. For local runs, the evaluator can auto-start the local Glass Bridge server and talk to it over `http://127.0.0.1:8000`, so the same command shape can later be reused remotely. ## OpenEnv-style server mode Start the local server: ```bash PYTHONPATH=src uvicorn openenv_glass_bridge.server.app:app --host 127.0.0.1 --port 8000 ``` Run the example client rollout against that server: ```bash PYTHONPATH=src python -m openenv_glass_bridge.examples.example_usage