Spaces:
Running
Running
Commit History
-
Vendor the necessary openra_rl_training + openra_env modules
Makes the bench self-contained for Python code — no external OpenRA-RL-Training / OpenRA-RL checkouts needed (addresses the other half of PR #12's concern, the correct way: faithful frozen copies, not stubs). Vendored verbatim (sha-verified identical to source) — only the modules the bench actually uses: openra_rl_training/ scenario.py — ScenarioDefinition, VALID_ACTOR_TYPES training/reward_funcs.py — DEFAULT_REWARD_WEIGHTS training/rust_env_pool.py — RustEnvPool training/minimap_renderer.py — render_minimap (terrain minimap) openra_env/ game_data.py — RA_UNITS / RA_BUILDINGS openra_env/__init__.py deliberately does NOT pull client/models — the only reference to those is example code inside a docstring; the bench never imports the legacy gRPC client at runtime. requirements.txt: add the now-needed deps (pydantic, pyyaml, pillow, numpy, matplotlib). VENDOR.md documents provenance + the re-vendor procedure. The Rust engine…9063f2b
