Skip to content

Hypergraph is more powerful. Transform unstructured text into structured knowledge with LLMs. Graphs, hypergraphs, and spatio-temporal extractions — with one command.

Merge PR #100: wire AutoDocument as YAML type and public export

yifanfeng978565598c459a
diff --cc hyperextract/utils/template_engine/factory.py
index 050f6ed,16f51a9..df20a03
--- a/hyperextract/utils/template_engine/factory.py
+++ b/hyperextract/utils/template_engine/factory.py
@@@ -1,8 -1,7 +1,8 @@@
"""Template Factory - Dynamically creates template instances from configuration.
- `create()` currently wires: model, list, set, graph, hypergraph,
- temporal_graph, spatial_graph, spatio_temporal_graph.
- Unknown `type` values raise ValueError (document is not wired here).
+ Supports AutoType generation for model, list, set, document, graph,
+ hypergraph, temporal_graph, spatial_graph, and spatio_temporal_graph.
++Unknown `type` values raise ValueError listing the allowed types.
"""
from pathlib import Path
diff --cc tests/template_engine/test_factory.py
index 15bc7a2,47b9d6b..f8f7f87
--- a/tests/template_engine/test_factory.py
+++ b/tests/template_engine/test_factory.py
@@@ -130,33 -127,48 +130,80 @@@ class TestTemplateFactoryCreateAllTypes
assert result.metadata.get("type") == "graph"
+class TestTemplateFactoryUnknownType:
+ """Unknown AutoType must fail closed with ValueError, not UnboundLocalError."""
+
+ def test_unknown_type_yaml_raises_valueerror(self, tmp_path, llm_client, embedder):
+ repo_root = Path(__file__).resolve().parents[2]
+ src = (repo_root / "hyperextract/templates/presets/general/base_model.yaml").read_text(
+ encoding="utf-8"
+ )
+ yaml_path = tmp_path / "not_a_type.yaml"
+ yaml_path.write_text(
+ src.replace("\ntype: model\n", "\ntype: not_a_type\n"),
+ encoding="utf-8",
+ )
+ raw_type = yaml.safe_load(yaml_path.read_text(encoding="utf-8"))["type"]
+ assert raw_type == "not_a_type"
+
+ # Parser Literal rejects unknown types at load; assignment reaches the
+ # factory match the same way an unbound type would after a looser schema.
+ cfg = Gallery.get("general/model").model_copy(deep=True)
+ cfg.type = raw_type
+
+ with pytest.raises(ValueError, match="not_a_type") as exc_info:
+ TemplateFactory.create(cfg, "en", llm_client, embedder)
+
+ assert not isinstance(exc_info.value, UnboundLocalError)
+ message = str(exc_info.value)
+ assert "Allowed types:" in message
+ assert "model" in message
+ assert "graph" in message
+ assert "hypergraph" in message
++
++
+ _DOCUMENT_YAML = """
+ language: en
+ name: document_fixture
+ type: document
+ tags: [test]
+ description: Minimal AutoDocument fixture (chunk corpus, no LLM extraction).
+ output:
+ description: Unused; AutoDocument stores raw chunks.
+ fields:
+ - name: content
+ type: str
+ description: Placeholder required by TemplateCfg.
+ guideline:
+ target: Unused; AutoDocument does not extract.
+ rules:
+ - Chunk text only; do not invent graph schema.
+ display:
+ label: '{content}'
+ """
+
+
+ class TestTemplateFactoryDocumentType:
+ def test_create_document_yaml_type(self, tmp_path, llm_client, embedder):
+ yaml_path = tmp_path / "document.yaml"
+ yaml_path.write_text(_DOCUMENT_YAML, encoding="utf-8")
+
+ from hyperextract.utils.template_engine import Template
+
+ result = Template.create(
+ str(yaml_path),
+ "en",
+ llm_client,
+ embedder,
+ )
+
+ from hyperextract import AutoDocument
+
+ assert isinstance(result, AutoDocument)
+ assert result.metadata["type"] == "document"
+
+ def test_public_import_autodocument(self):
+ from hyperextract import AutoDocument as Exported
+ from hyperextract.types import AutoDocument as Canonical
+
+ assert Exported is Canonical