Revert "feat: built-in default QRF k-point model with HuggingFace resolution"

.gitignore

@@ -15,15 +15,3 @@ notebooks/
 local_data/
 # Pi harness state
 .pi/
-
-# macOS
-.DS_Store
-
-# Built docs site (mkdocs output)
-site/
-
-# Pipeline run outputs
-goldilocks_output/
-
-# Stray planning scratch files
-plan-*.md

pyproject.toml

@@ -5,20 +5,10 @@ description = "Materials machine learning tools for recommending DFT calculation
 readme = "README.md"
 requires-python = ">=3.12"
 dependencies = [
-    "dscribe>=2.1",
-    "huggingface_hub>=0.23",
     "joblib>=1.5.3",
     "matminer>=0.10.1",
     "numpy>=2.4.4",
-    "pandas>=2.0",
     "pymatgen>=2026.3.23",
-    "sklearn-quantile>=0.1.1",
-    # The default k-points model (QRF + CGCNN metallicity) is a core capability,
-    # so its ML stack ships in the base install (decision: stfc/goldilocks-core#32).
-    # torch-geometric requires torch; on GPU hosts install the matching CUDA wheel
-    # of torch first, then sync.
-    "torch>=2.0",
-    "torch-geometric>=2.5",
 ]
 
 [project.scripts]

src/goldilocks_core/advisors/__init__.py

@@ -1,7 +1,6 @@
 """Advisor interfaces for calculation recommendations."""
 
-from .kdistance_advisor import default_kmesh_advisor as default_kmesh_advisor
-from .kindex_advisor import advise_kpoints as advise_kpoints
-from .kindex_advisor import ml_kmesh_advisor as ml_kmesh_advisor
+from .kmesh_advisor import advise_kpoints as advise_kpoints
+from .kmesh_advisor import ml_kmesh_advisor as ml_kmesh_advisor
 
-__all__ = ["advise_kpoints", "default_kmesh_advisor", "ml_kmesh_advisor"]
+__all__ = ["advise_kpoints", "ml_kmesh_advisor"]

src/goldilocks_core/cli/core.py

@@ -14,7 +14,7 @@
     CoreResult,
     ModelSpec,
 )
-from goldilocks_core.jobs import Pipeline, default_pipeline, run_core_job
+from goldilocks_core.jobs import Pipeline, run_core_job
 from goldilocks_core.pseudo.pp_registry import load_pseudo_metadata
 
 
@@ -160,29 +160,20 @@ def _request_from_args(args: argparse.Namespace) -> CoreJobRequest:
 
 
 def _pipeline_from_args(args: argparse.Namespace) -> Pipeline | None:
-    """Build the pipeline for the requested k-point backend.
-
-    Precedence: an explicit ``--model`` (local CSLR k-index model) wins; an
-    explicit k-point hint (``--k-grid``/``--k-spacing``) resolves from advice
-    without loading any model; otherwise the built-in default QRF pipeline runs
-    (with heuristic fallback).
-    """
-    if args.model is not None:
-        spec = ModelSpec(
-            name=args.model_name,
-            version=args.model_version,
-            model_type="random_forest",
-            target="k_index",
-            feature_set="cslr",
-            source="local",
-            location=args.model,
-        )
-        return Pipeline(kmesh=ml_kmesh_advisor(spec))
-
-    if args.k_grid is not None or args.k_spacing is not None:
+    """Build a custom pipeline from CLI backend options."""
+    if args.model is None:
         return None
 
-    return default_pipeline()
+    spec = ModelSpec(
+        name=args.model_name,
+        version=args.model_version,
+        model_type="random_forest",
+        target="k_index",
+        feature_set="cslr",
+        source="local",
+        location=args.model,
+    )
+    return Pipeline(kmesh=ml_kmesh_advisor(spec))
 
 
 def _parse_optional_bool(value: str | None) -> bool | None:

src/goldilocks_core/jobs.py

@@ -56,20 +56,6 @@ class Pipeline:
     bundle: BundleStage = write_bundle_directory
 
 
-def default_pipeline() -> Pipeline:
-    """Return a Pipeline whose Kmesh stage uses the built-in default ML model.
-
-    Unlike the bare ``Pipeline()`` (whose Kmesh stage is the spacing heuristic),
-    this resolves the default QRF k-point model and predicts k-points out of the
-    box, degrading to heuristic advice when the model or its dependencies are
-    unavailable. This is the pipeline the CLI uses when no ``--model`` and no
-    explicit k-point hint are given.
-    """
-    from goldilocks_core.advisors import default_kmesh_advisor
-
-    return Pipeline(kmesh=default_kmesh_advisor())
-
-
 def run_core_job(
     request: CoreJobRequest,
     *,