Merge branch 'main' into checkpoint-policy-refactor-v2

Author: relh

.github/actions/setup-environment/action.yml

@@ -54,7 +54,7 @@ runs:
     - name: Set up nim
       shell: bash
       run: |
-        curl -L -o nimby https://github.com/treeform/nimby/releases/download/0.1.6/nimby-Linux-X64
+        curl -L -o nimby https://github.com/treeform/nimby/releases/latest/download/nimby-Linux-X64
         chmod +x nimby
         ./nimby use 2.2.6
         echo "$HOME/.nimby/nim/bin" >> "$GITHUB_PATH"

.importlinter

@@ -18,7 +18,6 @@ root_packages =
     metta.cogworks
     metta.adaptive
     metta.map
-    metta.utils
     metta.common
     metta.app_backend
     cogames
@@ -41,7 +40,6 @@ layers =
     metta.cogworks
     metta.adaptive
     metta.map
-    metta.utils
     metta.common
 # Known violations to fix over time (remove as they get fixed)
 ignore_imports =
@@ -50,8 +48,6 @@ ignore_imports =
     metta.gridworks.** -> metta.tools.**
     metta.rl.** -> metta.tools.**
     metta.rl.** -> metta.sim.**
-    metta.utils.** -> metta.adaptive.**
-    metta.common.** -> metta.utils.**
     metta.common.** -> metta.rl.**
 
 [importlinter:contract:cogames-independence]
@@ -102,5 +98,4 @@ forbidden_modules =
 ignore_imports =
     metta.common.** -> metta.common.**
     # Tech debt - remove as violations are fixed:
-    metta.common.** -> metta.utils.**
     metta.common.** -> metta.rl.**

README.md

@@ -154,7 +154,6 @@ graph TD
     cogworks[cogworks]
     adaptive[adaptive]
     map[map]
-    utils[utils]
     common[common]
     tools --> gridworks
     gridworks --> setup
@@ -164,8 +163,7 @@ graph TD
     sweep --> cogworks
     cogworks --> adaptive
     adaptive --> map
-    map --> utils
-    utils --> common
+    map --> common
 ```
 
 <!-- FOLDER_DIAGRAM_END -->

common/src/metta/common/tool/run_tool.py

@@ -26,7 +26,7 @@
 from rich.console import Console
 from typing_extensions import TypeVar
 
-import metta.utils.torch_init  # noqa: F401
+import metta.rl.torch_init  # noqa: F401
 from metta.common.tool import Tool
 from metta.common.tool.recipe_registry import recipe_registry
 from metta.common.tool.schema import get_pydantic_field_info

metta/utils/live_run_monitor.py metta/adaptive/live_run_monitor.pymetta/utils/live_run_monitor.py renamed to metta/adaptive/live_run_monitor.py

@@ -15,12 +15,12 @@
 - Default: fetch 50 runs, display 10 most recent
 
 Usage:
-    ./metta/util/live_run_monitor.py --group my_group_name
-    ./metta/util/live_run_monitor.py --name-filter "axel.*"
-    ./metta/util/live_run_monitor.py --group my_group --name-filter "experiment.*"
-    ./metta/util/live_run_monitor.py --refresh 15 --entity myteam --project myproject
-    ./metta/util/live_run_monitor.py --fetch-limit 100 --display-limit 20
-    ./metta/util/live_run_monitor.py  # Monitor last 10 runs (fetch 50, display 10)
+    metta run-monitor --group my_group_name
+    metta run-monitor --name-filter "axel.*"
+    metta run-monitor --group my_group --name-filter "experiment.*"
+    metta run-monitor --refresh 15 --entity myteam --project myproject
+    metta run-monitor --fetch-limit 100 --display-limit 20
+    metta run-monitor  # Monitor last 10 runs (fetch 50, display 10)
 """
 
 import logging

metta/utils/torch_init.py metta/rl/torch_init.pymetta/utils/torch_init.py renamed to metta/rl/torch_init.py

@@ -35,7 +35,6 @@
 from metta.setup.tools.test_runner.test_nim import app as nim_test_runner_app
 from metta.setup.tools.test_runner.test_python import app as python_test_runner_app
 from metta.setup.utils import debug, error, info, success, warning
-from metta.utils.live_run_monitor import app as run_monitor_app
 
 if TYPE_CHECKING:
     from metta.setup.registry import SetupModule
@@ -718,7 +717,23 @@ def cmd_gridworks(ctx: typer.Context):
     subprocess.run(cmd, cwd=get_repo_root(), check=False)
 
 
-app.add_typer(run_monitor_app, name="run-monitor", help="Monitor training runs.")
+@app.command(
+    name="run-monitor",
+    help="Monitor training runs.",
+    context_settings={"allow_extra_args": True, "ignore_unknown_options": True},
+    add_help_option=False,
+)
+def cmd_run_monitor(ctx: typer.Context) -> None:
+    """Launch the live run monitor CLI without importing it at module load time."""
+    from metta.adaptive import live_run_monitor
+
+    live_run_monitor.app(
+        prog_name="metta run-monitor",
+        standalone_mode=False,
+        args=list(ctx.args),
+    )
+
+
 app.add_typer(observatory_app, name="observatory")
 app.add_typer(book_app, name="book")
 app.add_typer(codebase_app, name="codebase")

metta/setup/metta_cli.py

@@ -20,7 +20,7 @@
 )
 from torch.utils.data import DataLoader
 
-from metta.utils.torch_init import enable_determinism
+from metta.rl.torch_init import enable_determinism
 
 # Globals used by optional Axons parity probe
 AXONS_PARITY_PROBE: int = 0

metta/utils/__init__.py

@@ -109,7 +109,7 @@ def set_tf32_precision(mode: str) -> None:
 def configure_tf32_precision() -> None:
     """Ensure TF32 fast paths are enabled."""
     try:
-        from metta.utils.torch_init import configure_torch_globally
+        from metta.rl.torch_init import configure_torch_globally
 
         configure_torch_globally()
     except ImportError: