Merge pull request #8 from alea-institute/v0.1.5

merge V0.1.5 onto master

CHANGES.md

@@ -1,3 +1,7 @@
+Version 0.1.5 (2024-11-08)
+---------------------------
+* Adding support for LLM-backed (decoder) search, e.g., via OpenAI, Anthropic, VLLM, Together
+
 Version 0.1.4 (2024-09-04)
 ---------------------------
 * Add prefix search for typeahead/search bars (with optional trie-based search)

README.md

@@ -18,6 +18,9 @@ SOLI is an open, CC-BY licensed standard designed to represent universal element
 - Access detailed information about each class, including labels, definitions, and examples
 - Convert classes to OWL XML or Markdown format
 
+## Changelog
+The changelog can be found at [CHANGES.md](CHANGES.md).
+
 ## Installation
 
 You can install the SOLI Python library using pip:
@@ -58,6 +61,26 @@ for area in areas_of_law:
     print(area.label)
 ```
 
+## Searching with an LLM
+
+```python
+# Search with an LLM
+async def search_example():
+    for result in await soli.parallel_search_by_llm(
+        "redline lease agreement",
+        search_sets=[
+            soli.get_areas_of_law(max_depth=1),
+            soli.get_player_actors(max_depth=2),
+        ],
+    ):
+        print(result)
+
+import asyncio
+asyncio.run(search_example())
+```
+
+LLM search uses the `alea_llm_client` to provide abstraction across multiple APIs and providers.
+
 ## Documentation
 
 For more detailed information about using the SOLI Python library, please refer to our [full documentation](https://soli-python.readthedocs.io/).

docker/ubuntu2204-install/Dockerfile

@@ -1,7 +1,7 @@
 FROM ubuntu:22.04
 
 # define package version
-ARG SOLI_VERSION=0.1.4
+ARG SOLI_VERSION=0.1.5
 
 # Avoid prompts from apt
 ENV DEBIAN_FRONTEND=noninteractive

docker/ubuntu2404-install/Dockerfile

@@ -1,7 +1,7 @@
 FROM ubuntu:24.04
 
 # define package version
-ARG SOLI_VERSION=0.1.4
+ARG SOLI_VERSION=0.1.5
 
 # Avoid prompts from apt
 ENV DEBIAN_FRONTEND=noninteractive

docs/conf.py

@@ -25,8 +25,8 @@
 project = "soli-python"
 copyright = "2024, ALEA Institute"
 author = "ALEA Institute (https://aleainstitute.ai)"
-release = "0.1.4"
-version = "0.1.4"
+release = "0.1.5"
+version = "0.1.5"
 master_doc = "index"
 language = "en"
 

examples/llm_search.py

@@ -0,0 +1,55 @@
+import asyncio
+import time
+from pathlib import Path
+from soli import SOLI
+from alea_llm_client import OpenAIModel, AnthropicModel
+
+
+async def main():
+    #  text to label/classify
+    example_text = "review and revise license agreement"
+
+    # set at initialization
+    g = SOLI(llm=OpenAIModel(model="gpt-4o"))
+
+    print("gpt-4o results:")
+    for x in await g.parallel_search_by_llm(
+        example_text,
+    ):
+        print(x)
+
+    # use a small llama model for area of law
+    TOGETHER_API_KEY = (Path.home() / ".alea" / "keys" / "together").read_text().strip()
+    g.llm = OpenAIModel(
+        endpoint="https://api.together.xyz",
+        model="meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo",
+        api_key=TOGETHER_API_KEY,
+    )
+
+    print("\n\nmeta-llama/Llama-3.2-3B-Instruct-Turbo results:")
+    for x in await g.search_by_llm(
+        example_text,
+        search_set=g.get_areas_of_law(max_depth=1),
+    ):
+        print(x)
+
+    # override via property
+    g.llm = AnthropicModel(model="claude-3-5-haiku-20241022")
+
+    # search specific branches
+    print("\n\nclaude-3-5-haiku-20241022 results:")
+    for x in await g.parallel_search_by_llm(
+        example_text,
+        search_sets=[
+            g.get_areas_of_law(max_depth=2),
+            g.get_document_artifacts(max_depth=2),
+            g.get_player_actors(max_depth=3),
+        ],
+    ):
+        print(x)
+
+
+if __name__ == "__main__":
+    t0 = time.time()
+    asyncio.run(main())
+    print(time.time() - t0)

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "soli-python"
-version = "0.1.4"
+version = "0.1.5"
 description = "Python library for SOLI, the Standard for Open Legal Information"
 authors = ["ALEA Institute <[email protected]>"]
 license = "MIT"
@@ -39,9 +39,9 @@ python = ">=3.10,<4.0.0"
 pydantic = "^2.8.2"
 lxml = "^5.2.2"
 httpx = "^0.27.2"
-rapidfuzz = {version = "^3.9.7", optional = true}
+rapidfuzz = {version = "^3.10.0", optional = true}
 marisa-trie = {version = "^1.2.0", optional = true}
-
+alea-llm-client = {version = "^0.1.1", optional = true}
 
 [tool.poetry.group.dev.dependencies]
 types-lxml = "^2024.8.7"
@@ -62,10 +62,11 @@ sphinx-plausible = "^0.1.2"
 [tool.poetry.group.search.dependencies]
 rapidfuzz = "^3.9.7"
 marisa-trie = "^1.2.0"
+alea-llm-client = "^0.1.1"
 
 # extras
 [tool.poetry.extras]
-search = ["rapidfuzz", "marisa-trie"]
+search = ["rapidfuzz", "marisa-trie", "alea-llm-client"]
 
 
 [build-system]

soli/__init__.py

@@ -5,7 +5,7 @@
 # SPDX-License-Identifier: MIT
 # (c) 2024 ALEA Institute.
 
-__version__ = "0.1.4"
+__version__ = "0.1.5"
 __author__ = "ALEA Institute"
 __license__ = "MIT"
 __description__ = "Python library for SOLI, the Standard for Open Legal Information"