Dev (#13)

* docs: add authorization methods to README for GigaChat usage (#1) * ci: add workflow for dev branch (#2) * docs: use monthly downloads counter (#3) * docs: Readme update * feat: support loading prompts from github repos (#8) * docs: use monthly downloads counter * feat: support loading prompts from github repos --------- Co-authored-by: Dmitry Labazkin <[email protected]> * Support few-shots from class description (#10) * docs: use monthly downloads counter * few_shot_examples можно задавать через pydantic схему * feat: add pydantic schema support for few_shot_examples --------- Co-authored-by: Dmitry Labazkin <[email protected]> Co-authored-by: NIK-TIGER-BILL <[email protected]> * chore: minor version up (#14) --------- Co-authored-by: Dmitry Labazkin <[email protected]> Co-authored-by: NIK-TIGER-BILL <[email protected]>
ai-forever · Nov 26, 2024 · 1d57eb3 · 1d57eb3
1 parent 1a8364a
commit 1d57eb3
Show file tree

Hide file tree

Showing 9 changed files with 260 additions and 14 deletions.
diff --git a/.github/workflows/check_diffs.yml b/.github/workflows/check_diffs.yml
@@ -3,7 +3,9 @@ name: CI
 
 on:
   push:
-    branches: [master]
+    branches:
+      - master
+      - dev
   pull_request:
 
 # If another push to the same PR or branch happens while this workflow is still running,

diff --git a/libs/gigachat/README.md b/libs/gigachat/README.md
@@ -10,7 +10,45 @@ This is a library integration with [GigaChat](https://giga.chat/).
 [![GitHub Open Issues](https://img.shields.io/github/issues-raw/ai-forever/langchain-gigachat)](https://github.com/ai-forever/langchain-gigachat/issues)
 
 ## Installation
-
 ```bash
 pip install -U langchain-gigachat
-```
+```
+
+## Quickstart
+Follow these simple steps to get up and running quickly.
+
+### Installation
+To install the package use following command:
+```shell
+pip install -U langchain-gigachat
+```
+
+### Initialization
+
+To initialize chat model:
+```python
+from langchain_gigachat.chat_models import GigaChat
+
+giga = GigaChat(credentials="YOUR_AUTHORIZATION_KEY", verify_ssl_certs=False)
+```
+
+To initialize embeddings:
+
+```python
+from langchain_gigachat.embeddings import GigaChatEmbeddings
+
+embedding = GigaChatEmbeddings(
+    credentials="YOUR_AUTHORIZATION_KEY",
+    verify_ssl_certs=False
+)
+```
+
+### Usage
+
+Use the GigaChat object to generate responses:
+
+```python
+print(giga.invoke("Hello, world!"))
+```
+
+Now you can use the GigaChat object with LangChain's standard primitives to create LLM-applications.
diff --git a/libs/gigachat/langchain_gigachat/chat_models/gigachat.py b/libs/gigachat/langchain_gigachat/chat_models/gigachat.py
@@ -314,14 +314,43 @@ def trim_content_to_stop_sequence(
 class GigaChat(_BaseGigaChat, BaseChatModel):
     """`GigaChat` large language models API.
 
-    To use, you should pass login and password to access GigaChat API or use token.
+    To use, provide credentials via token, login and password,
+    or mTLS for secure access to the GigaChat API.
 
-    Example:
+    Example Usage:
         .. code-block:: python
 
             from langchain_community.chat_models import GigaChat
 
-            giga = GigaChat(credentials=..., scope=..., verify_ssl_certs=False)
+            # Authorization with Token
+            # (obtainable in the personal cabinet under Authorization Data):
+            giga = GigaChat(credentials="YOUR_TOKEN")
+
+            # Personal Space:
+            giga = GigaChat(credentials="YOUR_TOKEN", scope="GIGACHAT_API_PERS")
+
+            # Corporate Space:
+            giga = GigaChat(credentials="YOUR_TOKEN", scope="GIGACHAT_API_CORP")
+
+            # Authorization with Login and Password:
+            giga = GigaChat(
+                base_url="https://gigachat.devices.sberbank.ru/api/v1",
+                user="YOUR_USERNAME",
+                password="YOUR_PASSWORD",
+            )
+
+            # Mutual Authentication via TLS (mTLS):
+            giga = GigaChat(
+                base_url="https://gigachat.devices.sberbank.ru/api/v1",
+                ca_bundle_file="certs/ca.pem",  # chain_pem.txt
+                cert_file="certs/tls.pem",       # published_pem.txt
+                key_file="certs/tls.key",
+                key_file_password="YOUR_KEY_PASSWORD",
+            )
+
+            # Authorization with Temporary Token:
+            giga = GigaChat(access_token="YOUR_TEMPORARY_TOKEN")
+
     """
 
     def _build_payload(self, messages: List[BaseMessage], **kwargs: Any) -> gm.Chat:

diff --git a/libs/gigachat/langchain_gigachat/tools/load_prompt.py b/libs/gigachat/langchain_gigachat/tools/load_prompt.py
@@ -0,0 +1,67 @@
+"""Utilities for loading templates from gigachain
+github-based hub or other extenal sources."""
+
+import os
+import re
+import tempfile
+from pathlib import Path, PurePosixPath
+from typing import Any, Callable, Optional, Set, TypeVar, Union
+from urllib.parse import urljoin
+
+import requests
+from langchain_core.prompts.base import BasePromptTemplate
+from langchain_core.prompts.loading import _load_prompt_from_file
+
+DEFAULT_REF = os.environ.get("GIGACHAIN_HUB_DEFAULT_REF", "master")
+URL_BASE = os.environ.get(
+    "GIGACHAIN_HUB_DEFAULT_REF",
+    "https://raw.githubusercontent.com/ai-forever/gigachain/{ref}/hub/",
+)
+HUB_PATH_RE = re.compile(r"lc(?P<ref>@[^:]+)?://(?P<path>.*)")
+
+T = TypeVar("T")
+
+
+def _load_from_giga_hub(
+    path: Union[str, Path],
+    loader: Callable[[str], T],
+    valid_prefix: str,
+    valid_suffixes: Set[str],
+    **kwargs: Any,
+) -> Optional[T]:
+    """Load configuration from hub.  Returns None if path is not a hub path."""
+    if not isinstance(path, str) or not (match := HUB_PATH_RE.match(path)):
+        return None
+    ref, remote_path_str = match.groups()
+    ref = ref[1:] if ref else DEFAULT_REF
+    remote_path = Path(remote_path_str)
+    if remote_path.parts[0] != valid_prefix:
+        return None
+    if remote_path.suffix[1:] not in valid_suffixes:
+        raise ValueError(f"Unsupported file type, must be one of {valid_suffixes}.")
+
+    # Using Path with URLs is not recommended, because on Windows
+    # the backslash is used as the path separator, which can cause issues
+    # when working with URLs that use forward slashes as the path separator.
+    # Instead, use PurePosixPath to ensure that forward slashes are used as the
+    # path separator, regardless of the operating system.
+    full_url = urljoin(URL_BASE.format(ref=ref), PurePosixPath(remote_path).__str__())
+
+    r = requests.get(full_url, timeout=5)
+    if r.status_code != 200:
+        raise ValueError(f"Could not find file at {full_url}")
+    with tempfile.TemporaryDirectory() as tmpdirname:
+        file = Path(tmpdirname) / remote_path.name
+        with open(file, "wb") as f:
+            f.write(r.content)
+        return loader(str(file), **kwargs)
+
+
+def load_from_giga_hub(path: Union[str, Path]) -> BasePromptTemplate:
+    """Unified method for loading a prompt from GigaChain repo or local fs."""
+    if hub_result := _load_from_giga_hub(
+        path, _load_prompt_from_file, "prompts", {"py", "json", "yaml"}
+    ):
+        return hub_result
+    else:
+        raise ValueError("Prompt not found in GigaChain Hub.")
diff --git a/libs/gigachat/langchain_gigachat/utils/function_calling.py b/libs/gigachat/langchain_gigachat/utils/function_calling.py
@@ -365,6 +365,11 @@ def convert_pydantic_to_gigachat_function(
             "Incorrect function or tool description. Description is required."
         )
 
+    if few_shot_examples is None and hasattr(model, "few_shot_examples"):
+        few_shot_examples_attr = getattr(model, "few_shot_examples")
+        if inspect.isfunction(few_shot_examples_attr):
+            few_shot_examples = few_shot_examples_attr()
+
     return GigaFunctionDescription(
         name=name or title,
         description=description,

diff --git a/libs/gigachat/poetry.lock b/libs/gigachat/poetry.lock
diff --git a/libs/gigachat/pyproject.toml b/libs/gigachat/pyproject.toml
@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "langchain-gigachat"
-version = "0.3.0"
+version = "0.3.1"
 description = "An integration package connecting GigaChat and LangChain"
 authors = []
 readme = "README.md"
@@ -13,7 +13,8 @@ license = "MIT"
 [tool.poetry.dependencies]
 python = ">=3.9,<4.0"
 langchain-core = "^0.3"
-gigachat = "^0.1.35"
+gigachat = "^0.1.36"
+types-requests = "^2.32"
 
 [tool.poetry.group.dev]
 optional = true
@@ -41,6 +42,7 @@ pytest = "^8.3.3"
 pytest-cov = "^5.0.0"
 pytest-asyncio = "^0.24.0"
 pytest-mock = "^3.14.0"
+requests_mock = "^1.12.1"
 
 [build-system]
 requires = ["poetry-core>=1.0.0"]

diff --git a/libs/gigachat/tests/unit_tests/test_gigachat.py b/libs/gigachat/tests/unit_tests/test_gigachat.py
@@ -31,7 +31,7 @@
     _convert_dict_to_message,
     _convert_message_to_dict,
 )
-from langchain_gigachat.tools.giga_tool import giga_tool
+from langchain_gigachat.tools.giga_tool import FewShotExamples, giga_tool
 from tests.unit_tests.stubs import FakeAsyncCallbackHandler, FakeCallbackHandler
 
 
@@ -334,3 +334,44 @@ def test_gigachat_bind_gigatool() -> None:
         "required": ["status", "message"],
         "type": "object",
     }
+
+
+class SomeResult(BaseModel):
+    """My desc"""
+
+    @staticmethod
+    def few_shot_examples() -> FewShotExamples:
+        return [
+            {
+                "request": "request example",
+                "params": {"is_valid": 1, "description": "correct message"},
+            }
+        ]
+
+    value: int = Field(description="some value")
+    description: str = Field(description="some descriptin")
+
+
+def test_structured_output() -> None:
+    llm = GigaChat().with_structured_output(SomeResult)
+    assert llm.steps[0].kwargs["function_call"] == {"name": "SomeResult"}  # type: ignore[attr-defined]
+    assert llm.steps[0].kwargs["tools"][0]["function"] == {  # type: ignore[attr-defined]
+        "name": "SomeResult",
+        "description": "My desc",
+        "parameters": {
+            "description": "My desc",
+            "properties": {
+                "value": {"description": "some value", "type": "integer"},
+                "description": {"description": "some descriptin", "type": "string"},
+            },
+            "required": ["value", "description"],
+            "type": "object",
+        },
+        "return_parameters": None,
+        "few_shot_examples": [
+            {
+                "request": "request example",
+                "params": {"is_valid": 1, "description": "correct message"},
+            }
+        ],
+    }
diff --git a/libs/gigachat/tests/unit_tests/test_utils.py b/libs/gigachat/tests/unit_tests/test_utils.py
@@ -0,0 +1,31 @@
+from typing import Generator
+
+import pytest
+import requests_mock
+from langchain_core.prompts.prompt import PromptTemplate
+
+from langchain_gigachat.tools.load_prompt import load_from_giga_hub
+
+
+@pytest.fixture
+def mock_requests_get() -> Generator:
+    with requests_mock.Mocker() as mocker:
+        mocker.get(
+            "https://raw.githubusercontent.com/ai-forever/gigachain/master/hub/prompts/entertainment/meditation.yaml",
+            text=(
+                "input_variables: [background, topic]\n"
+                "output_parser: null\n"
+                "template: 'Create mediation for {topic} with {background}'\n"
+                "template_format: f-string\n"
+                "_type: prompt"
+            ),
+        )
+        yield mocker
+
+
+def test__load_from_giga_hub(mock_requests_get: Generator) -> None:
+    template = load_from_giga_hub("lc://prompts/entertainment/meditation.yaml")
+    assert isinstance(template, PromptTemplate)
+    assert template.template == "Create mediation for {topic} with {background}"
+    assert "background" in template.input_variables
+    assert "topic" in template.input_variables