Skip to content

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

feat: wire AutoDocument as YAML type and public export

Frankie-Xu36d5af4edfbd
diff --git a/docs/en/python/index.md b/docs/en/python/index.md
index 2a602a9..45a9c3b 100644
--- a/docs/en/python/index.md
+++ b/docs/en/python/index.md
@@ -68,7 +68,7 @@ ka = Template.create("method/light_rag")
### Auto-Types
-Eight data structure types for different extraction needs:
+Eight data structure types for different extraction needs, plus `AutoDocument` for chunk-only corpora (`type: document` in YAML; no LLM extraction).
| Class | Use Case |
|-------|----------|
diff --git a/docs/zh/python/index.md b/docs/zh/python/index.md
index f344083..120fcf0 100644
--- a/docs/zh/python/index.md
+++ b/docs/zh/python/index.md
@@ -68,7 +68,7 @@ ka = Template.create("method/light_rag")
### 自动类型
-8 种用于不同提取需求的数据结构类型:
+8 种用于不同提取需求的数据结构类型,另有 `AutoDocument` 用于只切块、不走 LLM 抽取的语料(YAML `type: document`)。
| 类 | 用例 |
|-------|----------|
diff --git a/hyperextract/__init__.py b/hyperextract/__init__.py
index e8d746d..8f146ba 100644
--- a/hyperextract/__init__.py
+++ b/hyperextract/__init__.py
@@ -25,6 +25,7 @@ Usage:
from importlib.metadata import version
from .types import (
+ AutoDocument,
AutoGraph,
AutoHypergraph,
AutoList,
@@ -50,6 +51,8 @@ __author__ = "Yifan Feng"
__email__ = "[email protected]"
__all__ = [
+ # Corpus type
+ "AutoDocument",
# Graph types
"AutoGraph",
"AutoHypergraph",
diff --git a/hyperextract/utils/template_engine/factory.py b/hyperextract/utils/template_engine/factory.py
index 77e43ca..16f51a9 100644
--- a/hyperextract/utils/template_engine/factory.py
+++ b/hyperextract/utils/template_engine/factory.py
@@ -1,6 +1,7 @@
"""Template Factory - Dynamically creates template instances from configuration.
-Supports all 8 AutoType dynamic generation.
+Supports AutoType generation for model, list, set, document, graph,
+hypergraph, temporal_graph, spatial_graph, and spatio_temporal_graph.
"""
from pathlib import Path
@@ -21,6 +22,7 @@ from .parsers import (
if TYPE_CHECKING:
from hyperextract.types import (
+ AutoDocument,
AutoGraph,
AutoHypergraph,
AutoList,
@@ -375,6 +377,24 @@ class TemplateFactory:
**options,
)
+ @classmethod
+ def create_document(
+ cls,
+ config: TemplateCfg,
+ llm_client: BaseChatModel,
+ embedder: Embeddings,
+ **kwargs,
+ ) -> "AutoDocument":
+ """Create AutoDocument template (chunk corpus, no LLM extraction)."""
+ from hyperextract.types import AutoDocument
+
+ options = parse_option(config.options, config.type, override=kwargs)
+ return AutoDocument(
+ llm_client=llm_client,
+ embedder=embedder,
+ **options,
+ )
+
@classmethod
def create(
cls,
@@ -473,6 +493,10 @@ class TemplateFactory:
template = cls.create_list(template_cfg, llm_client, embedder, **kwargs)
case "set":
template = cls.create_set(template_cfg, llm_client, embedder, **kwargs)
+ case "document":
+ template = cls.create_document(
+ template_cfg, llm_client, embedder, **kwargs
+ )
case "graph":
template = cls.create_graph(
template_cfg, llm_client, embedder, **kwargs
diff --git a/hyperextract/utils/template_engine/parsers/loader.py b/hyperextract/utils/template_engine/parsers/loader.py
index 09d49a4..fd163cd 100644
--- a/hyperextract/utils/template_engine/parsers/loader.py
+++ b/hyperextract/utils/template_engine/parsers/loader.py
@@ -92,7 +92,7 @@ def _localize_output(
autotype: VALID_AUTOTYPES,
) -> NaiveOutputSchema | GraphOutputSchema:
"""Localize output configuration."""
- if autotype in ("model", "list", "set"):
+ if autotype in ("model", "list", "set", "document"):
return _localize_naive_output(output, language)
return GraphOutputSchema(
description=_localize_data(output.description, language),
@@ -107,7 +107,7 @@ def _localize_guideline(
autotype: VALID_AUTOTYPES,
) -> NaiveGuidelineSchema | GraphGuidelineSchema:
"""Localize guideline configuration."""
- if autotype in ("model", "list", "set"):
+ if autotype in ("model", "list", "set", "document"):
return NaiveGuidelineSchema(
target=_localize_data(guideline.target, language),
rules=_localize_data(guideline.rules, language),
diff --git a/hyperextract/utils/template_engine/parsers/options.py b/hyperextract/utils/template_engine/parsers/options.py
index 58e55e7..fcb2ec1 100644
--- a/hyperextract/utils/template_engine/parsers/options.py
+++ b/hyperextract/utils/template_engine/parsers/options.py
@@ -131,6 +131,9 @@ def parse_option(
if autotype in ("set",):
return _build_kwargs(options, SET_PARAMS, COMMON_PARAMS)
+ if autotype in ("document",):
+ return _build_kwargs(options, (), COMMON_PARAMS)
+
if autotype in ("graph", "hypergraph"):
return _build_kwargs(options, GRAPH_PARAMS, COMMON_PARAMS)
diff --git a/hyperextract/utils/template_engine/parsers/schemas/base.py b/hyperextract/utils/template_engine/parsers/schemas/base.py
index fa7a6d3..787d57e 100644
--- a/hyperextract/utils/template_engine/parsers/schemas/base.py
+++ b/hyperextract/utils/template_engine/parsers/schemas/base.py
@@ -14,6 +14,7 @@ VALID_AUTOTYPES = Literal[
"model",
"list",
"set",
+ "document",
"graph",
"hypergraph",
"temporal_graph",
diff --git a/tests/template_engine/test_factory.py b/tests/template_engine/test_factory.py
index fb79318..47b9d6b 100644
--- a/tests/template_engine/test_factory.py
+++ b/tests/template_engine/test_factory.py
@@ -125,3 +125,50 @@ class TestTemplateFactoryCreateAllTypes:
assert isinstance(result, AutoGraph)
assert result.metadata.get("type") == "graph"
+
+
+_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