python-gfm 0.1.2__tar.gz → 0.1.3__tar.gz

{python_gfm-0.1.2 → python_gfm-0.1.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-gfm
-Version: 0.1.2
+Version: 0.1.3
 Summary: An AI copilot for graph data and models (Under active development).
 Author-email: BUAA SKLCCSE <your.email@example.com>
 License: Apache-2.0

{python_gfm-0.1.2 → python_gfm-0.1.3}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "python-gfm"
-version = "0.1.2"
+version = "0.1.3"
 authors = [
   { name = "BUAA SKLCCSE", email = "your.email@example.com" },
 ]

{python_gfm-0.1.2 → python_gfm-0.1.3}/python_gfm.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-gfm
-Version: 0.1.2
+Version: 0.1.3
 Summary: An AI copilot for graph data and models (Under active development).
 Author-email: BUAA SKLCCSE <your.email@example.com>
 License: Apache-2.0

{python_gfm-0.1.2 → python_gfm-0.1.3}/python_gfm.egg-info/SOURCES.txt

@@ -45,7 +45,6 @@ src/pygfm/baseline_models/gcot/graphgpt/model/model_registry.py
 src/pygfm/baseline_models/gcot/graphgpt/model/monkey_patch_non_inplace.py
 src/pygfm/baseline_models/gcot/graphgpt/model/utils.py
 src/pygfm/baseline_models/gcot/graphgpt/model/graph_layers/__init__.py
-src/pygfm/baseline_models/gcot/graphgpt/model/graph_layers/bpe_simple_vocab_16e6.txt.gz
 src/pygfm/baseline_models/gcot/graphgpt/model/graph_layers/clip_graph.py
 src/pygfm/baseline_models/gcot/graphgpt/model/graph_layers/graph_transformer.py
 src/pygfm/baseline_models/gcot/graphgpt/model/graph_layers/mpnn.py
@@ -106,7 +105,6 @@ src/pygfm/baseline_models/graphgpt/model/model_registry.py
 src/pygfm/baseline_models/graphgpt/model/monkey_patch_non_inplace.py
 src/pygfm/baseline_models/graphgpt/model/utils.py
 src/pygfm/baseline_models/graphgpt/model/graph_layers/__init__.py
-src/pygfm/baseline_models/graphgpt/model/graph_layers/bpe_simple_vocab_16e6.txt.gz
 src/pygfm/baseline_models/graphgpt/model/graph_layers/clip_graph.py
 src/pygfm/baseline_models/graphgpt/model/graph_layers/graph_transformer.py
 src/pygfm/baseline_models/graphgpt/model/graph_layers/mpnn.py

{python_gfm-0.1.2 → python_gfm-0.1.3}/src/pygfm/public/utils/runtime.py

@@ -218,23 +218,27 @@ def _match_dataset_mapping_key(folder_name: str, dataset_mapping: dict) -> str |
     return None
 
 
-def load_all_datasets(data_root) -> list:
+def load_all_datasets(data_root: str, *, allow_pyg_download: bool = False) -> list:
     """
     Scan ``data_root`` for graphs; returns ``list[dict]`` compatible with ``{d['name']: d for d in all_raw}``.
 
     **Flat ``*.pt``**: single-file graphs like ``data_root/Cora.pt`` loaded to one ``Data``;
     names are matched **case-insensitively** to ``dataset_mapping`` (e.g. ``cora.pt`` -> ``Cora``).
 
-    **Subdirs**: ``<name>/data.pt`` or ``<name>/processed/data.pt`` before PyG download fallback.
+    **Subdirs**: ``<name>/data.pt`` or ``<name>/processed/data.pt``.
 
-    **Else**: use ``dataset_mapping`` (Planetoid / Amazon / TU / HGPROMPT).
+    **Else**:
+    - If ``allow_pyg_download=False`` (default), do **not** fall back to PyG datasets
+      (Planetoid/Amazon/TU). This avoids implicit downloads and enforces a uniform local
+      ``*.pt``/``data.pt`` format for both pretrain and finetune.
+    - If ``allow_pyg_download=True``, fall back to ``dataset_mapping`` (Planetoid / Amazon / TU / HGPROMPT).
     Use ``--datasets`` / ``--target`` names consistent with ``d['name']`` (e.g. ``Cora``).
     """
     datasets = []
     # Undirected + self-loops for robustness
     transform = T.Compose([T.ToUndirected(), T.AddSelfLoops()])
 
-    # When no data.pt: pick loader by name
+    # Optional PyG loader mapping (used only when allow_pyg_download=True)
     dataset_mapping = {
         "ACM": "HETEROGENEOUS",
         "DBLP": "HETEROGENEOUS",
@@ -381,27 +385,35 @@ def load_all_datasets(data_root) -> list:
                     else:
                         print(f"!! Warning: {canonical_name} has no valid edges after filter; skipped.")
 
-            elif dtype == Planetoid:
-                # Use on-disk folder ``name`` (Cora vs cora path case on Linux)
-                print(f">> Loading Planetoid: {canonical_name} (dir: {name})")
-                ds = Planetoid(root=data_root, name=name, transform=transform)
-                datasets.append({"name": canonical_name, "ds": ds})
-                loaded_canonical.add(canonical_name)
-                print(f"Successfully initialized: {canonical_name} (Nodes: {ds[0].num_nodes})")
-
-            elif dtype == Amazon:
-                print(f">> Loading Amazon: {canonical_name} (dir: {name})")
-                ds = Amazon(root=data_root, name=name, transform=transform)
-                datasets.append({"name": canonical_name, "ds": ds})
-                loaded_canonical.add(canonical_name)
-                print(f"Successfully initialized: {canonical_name} (Nodes: {ds[0].num_nodes})")
-
-            elif dtype == TUDataset:
-                print(f">> Loading TUDataset: {canonical_name} (dir: {name})")
-                ds = TUDataset(root=data_root, name=name, transform=transform)
-                datasets.append({"name": canonical_name, "ds": ds})
-                loaded_canonical.add(canonical_name)
-                print(f"Successfully initialized: {canonical_name} (Graphs: {len(ds)})")
+            elif dtype in (Planetoid, Amazon, TUDataset):
+                if not allow_pyg_download:
+                    print(
+                        f"!! Skip PyG fallback for {canonical_name} (dir: {name}). "
+                        f"Please provide a local export: "
+                        f"{os.path.join(data_root, canonical_name + '.pt')} "
+                        f"or {os.path.join(data_path, 'data.pt')}."
+                    )
+                    continue
+
+                if dtype == Planetoid:
+                    # Use on-disk folder ``name`` (Cora vs cora path case on Linux)
+                    print(f">> Loading Planetoid: {canonical_name} (dir: {name})")
+                    ds = Planetoid(root=data_root, name=name, transform=transform)
+                    datasets.append({"name": canonical_name, "ds": ds})
+                    loaded_canonical.add(canonical_name)
+                    print(f"Successfully initialized: {canonical_name} (Nodes: {ds[0].num_nodes})")
+                elif dtype == Amazon:
+                    print(f">> Loading Amazon: {canonical_name} (dir: {name})")
+                    ds = Amazon(root=data_root, name=name, transform=transform)
+                    datasets.append({"name": canonical_name, "ds": ds})
+                    loaded_canonical.add(canonical_name)
+                    print(f"Successfully initialized: {canonical_name} (Nodes: {ds[0].num_nodes})")
+                else:
+                    print(f">> Loading TUDataset: {canonical_name} (dir: {name})")
+                    ds = TUDataset(root=data_root, name=name, transform=transform)
+                    datasets.append({"name": canonical_name, "ds": ds})
+                    loaded_canonical.add(canonical_name)
+                    print(f"Successfully initialized: {canonical_name} (Graphs: {len(ds)})")
 
         except Exception as e:
             print(f"!! Failed to load {canonical_name} ({name}) due to error: {e}")
@@ -413,14 +425,16 @@ def load_all_datasets(data_root) -> list:
             subs = []
         print(
             f"!! Warning: no datasets loaded. data_root={data_root} | entries: {subs}."
-            f" Use flat Cora.pt; or Cora/data.pt, Cora/processed/data.pt;"
-            f" or standard Planetoid/Amazon layouts."
+            f" Provide flat Cora.pt; or Cora/data.pt, Cora/processed/data.pt."
+            + ("" if not allow_pyg_download else " Or enable PyG fallback with allow_pyg_download=True.")
         )
 
     return datasets
 
 
-def load_single_graph_dataset(data_root: str, dataset: str) -> tuple[Data, int]:
+def load_single_graph_dataset(
+    data_root: str, dataset: str, *, allow_pyg_download: bool = False
+) -> tuple[Data, int]:
     """
     Load **one** node-level graph under a baseline data root (same rules as :func:`load_all_datasets`):
 
@@ -433,7 +447,7 @@ def load_single_graph_dataset(data_root: str, dataset: str) -> tuple[Data, int]:
     :return: ``(data, num_classes)`` with single :class:`~torch_geometric.data.Data`.
     :raises ValueError: Not found, or multi-graph TU entry (use :func:`load_multi_graph_pyg_dataset` for graph classification).
     """
-    items = load_all_datasets(data_root)
+    items = load_all_datasets(data_root, allow_pyg_download=allow_pyg_download)
     key = dataset.strip()
     found_names = [d["name"] for d in items]
     for d in items:
@@ -504,20 +518,13 @@ def load_multi_graph_pyg_dataset(data_root: str, dataset: str):
 
 def load_single_graph_dataset_or_reddit(data_root: str, dataset: str) -> tuple[Data, int]:
     """
-    Same as :func:`load_single_graph_dataset`; if ``reddit`` has no ``.pt`` export,
-    fall back to PyG ``Reddit(root=data_root)`` (download or existing ``Reddit/raw``).
+    Same as :func:`load_single_graph_dataset`.
+
+    Note:
+    This helper used to fall back to PyG ``Reddit(root=data_root)`` if no local ``.pt`` export exists.
+    For a uniform, offline-friendly workflow, require local exports here as well.
     """
-    try:
-        return load_single_graph_dataset(data_root, dataset)
-    except ValueError:
-        if dataset.strip().casefold() != "reddit":
-            raise
-        from torch_geometric.datasets import Reddit
-
-        ds = Reddit(root=data_root)
-        data = ds[0]
-        num_classes = int(data.y.max().item()) + 1
-        return data, num_classes
+    return load_single_graph_dataset(data_root, dataset, allow_pyg_download=False)
 
 
 def early_stopping(