.env.example

# rl-studio — local env for the Modal GPU path. Copy to .env (gitignored).
#
# IMPORTANT: Modal auth is NOT here. Modal reads ~/.modal.toml (machine-level,
# set once with `modal token set` / `modal setup`). If `modal token current`
# works in your shell, the GPU path is already authenticated.
#
# Nothing in this file is required for the default run: Qwen2.5-0.5B is an open
# model and GSM8K is a public dataset, so no token is needed.

# Optional — only for GATED HuggingFace models. When set, it is forwarded into
# the Modal container as a Secret (see scripts/modal_grpo.py).
HF_TOKEN=

# Optional run knobs — override the config without editing it.
# RL_MODAL_GPU=A10G          # T4 | L4 | A10G | A100 | H100
# RL_MODAL_TIMEOUT_S=3600