Merge pull request #47 from CalculatedContent/agent/enforce-tmp-nanogpt-root

Enforce /tmp-only nanoGPT runtime paths

Author: charlesmartin14

baseline/nanogpt_one_head/README.md

@@ -230,8 +230,8 @@ cd baseline/nanogpt_one_head
 # Install this package and any missing dependencies into the active conda env.
 python -m pip install -e .
 
-# Keep all corpus caches and experiment outputs outside the git checkout.
-export RG_NANOGPT_ONE_HEAD_ROOT="$HOME/rg-nanogpt-one-head"
+# Keep all corpus caches and experiment outputs under /tmp.
+export RG_NANOGPT_ONE_HEAD_ROOT="/tmp/rg-nanogpt-one-head"
 
 # Allow unsupported individual MPS operations to fall back to CPU when needed.
 export PYTORCH_ENABLE_MPS_FALLBACK=1

baseline/nanogpt_one_head/src/rg_nanogpt_one_head/config.py

@@ -10,22 +10,22 @@
 import yaml
 
 SUPPORTED_OPTIMIZERS = ("sgd_momentum", "adamw", "muon")
-DEFAULT_ROOT = Path.home() / "rg-nanogpt-one-head"
+DEFAULT_ROOT = Path("/tmp/rg-nanogpt-one-head")
 
 
 def roots() -> dict[str, Path]:
-    root = Path(os.environ.get("RG_NANOGPT_ONE_HEAD_ROOT", DEFAULT_ROOT)).expanduser()
+    root = Path(os.environ.get("RG_NANOGPT_ONE_HEAD_ROOT", DEFAULT_ROOT))
     return {
         "root": root,
         "data": Path(
             os.environ.get("RG_NANOGPT_ONE_HEAD_DATA_ROOT", root / "data")
-        ).expanduser(),
+        ),
         "results": Path(
             os.environ.get("RG_NANOGPT_ONE_HEAD_RESULTS_ROOT", root / "results")
-        ).expanduser(),
+        ),
         "plots": Path(
             os.environ.get("RG_NANOGPT_ONE_HEAD_PLOTS_ROOT", root / "plots")
-        ).expanduser(),
+        ),
     }
 
 

baseline/nanogpt_one_head/src/rg_nanogpt_one_head/data.py

@@ -257,7 +257,8 @@ def prepare_fineweb_edu(
     except ImportError as exc:
         raise RuntimeError(
             "data preparation requires datasets and tiktoken; "
-            "run scripts/setup_mac.sh"
+            "install dependencies into the active conda environment with "
+            "`python -m pip install -e .`"
         ) from exc
 
     dataset_cfg = cfg["dataset"]

baseline/nanogpt_one_head/tests/test_paths.py

@@ -0,0 +1,72 @@
+from __future__ import annotations
+
+from pathlib import Path
+import sys
+
+
+EXPERIMENT_ROOT = Path(__file__).resolve().parents[1]
+REPO_ROOT = EXPERIMENT_ROOT.parents[1]
+sys.path.insert(0, str(EXPERIMENT_ROOT / "src"))
+
+from rg_nanogpt_one_head.config import DEFAULT_ROOT, roots
+
+
+ROOT_ENV_VARS = (
+    "RG_NANOGPT_ONE_HEAD_ROOT",
+    "RG_NANOGPT_ONE_HEAD_DATA_ROOT",
+    "RG_NANOGPT_ONE_HEAD_RESULTS_ROOT",
+    "RG_NANOGPT_ONE_HEAD_PLOTS_ROOT",
+)
+FORBIDDEN_HOME_TOKENS = (
+    "$HOME",
+    "${HOME}",
+    "Path.home(",
+    ".expanduser(",
+    "/home/",
+    "~/",
+)
+
+
+def test_default_root_is_tmp(monkeypatch):
+    for name in ROOT_ENV_VARS:
+        monkeypatch.delenv(name, raising=False)
+
+    assert DEFAULT_ROOT == Path("/tmp/rg-nanogpt-one-head")
+    resolved = roots()
+    assert resolved == {
+        "root": Path("/tmp/rg-nanogpt-one-head"),
+        "data": Path("/tmp/rg-nanogpt-one-head/data"),
+        "results": Path("/tmp/rg-nanogpt-one-head/results"),
+        "plots": Path("/tmp/rg-nanogpt-one-head/plots"),
+    }
+
+
+def test_nanogpt_sources_docs_and_notebooks_do_not_reference_home():
+    paths = [EXPERIMENT_ROOT / "README.md"]
+    paths.extend((EXPERIMENT_ROOT / "src").rglob("*.py"))
+    paths.extend((EXPERIMENT_ROOT / "notebooks").glob("*.ipynb"))
+    paths.extend((EXPERIMENT_ROOT / "configs").glob("*.yaml"))
+
+    violations = []
+    for path in paths:
+        text = path.read_text(encoding="utf-8")
+        for token in FORBIDDEN_HOME_TOKENS:
+            if token in text:
+                violations.append(f"{path.relative_to(REPO_ROOT)}: {token}")
+
+    assert not violations, "home-directory references found:\n" + "\n".join(violations)
+
+
+def test_repo_shell_scripts_never_reference_home():
+    violations = []
+    for path in REPO_ROOT.rglob("*.sh"):
+        text = path.read_text(encoding="utf-8")
+        for token in FORBIDDEN_HOME_TOKENS:
+            if token in text:
+                violations.append(f"{path.relative_to(REPO_ROOT)}: {token}")
+
+    assert not violations, "home-directory references found in shell scripts:\n" + "\n".join(violations)
+
+
+def test_removed_nanogpt_wrapper_scripts_do_not_return():
+    assert not (EXPERIMENT_ROOT / "scripts").exists()