fabricatio 0.2.3.dev0__tar.gz → 0.2.3.dev1__tar.gz

{fabricatio-0.2.3.dev0 → fabricatio-0.2.3.dev1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fabricatio
-Version: 0.2.3.dev0
+Version: 0.2.3.dev1
 Classifier: License :: OSI Approved :: MIT License
 Classifier: Programming Language :: Rust
 Classifier: Programming Language :: Python :: 3.12

{fabricatio-0.2.3.dev0 → fabricatio-0.2.3.dev1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "fabricatio"
-version = "0.2.3-dev0"
+version = "0.2.3-dev1"
 description = "A LLM multi-agent framework."
 readme = "README.md"
 license = { file = "LICENSE" }

fabricatio-0.2.3.dev1/python/fabricatio/capabilities/rag.py

@@ -0,0 +1,46 @@
+"""A module for the RAG (Retrieval Augmented Generation) model."""
+
+from typing import Any, Dict, List, Optional, Self, Union
+
+from fabricatio.config import configs
+from fabricatio.models.utils import MilvusData
+
+try:
+    from pymilvus import MilvusClient
+except ImportError as e:
+    raise RuntimeError("pymilvus is not installed. Have you installed `fabricatio[rag]` instead of `fabricatio`") from e
+from pydantic import BaseModel, ConfigDict, PrivateAttr
+
+
+class Rag(BaseModel):
+    """A class representing the RAG (Retrieval Augmented Generation) model."""
+
+    model_config = ConfigDict(use_attribute_docstrings=True)
+
+    _client: MilvusClient = PrivateAttr(
+        default=MilvusClient(
+            uri=configs.rag.milvus_uri.unicode_string(),
+            token=configs.rag.milvus_token.get_secret_value(),
+            timeout=configs.rag.milvus_timeout,
+        ),
+    )
+    _target_collection: Optional[str] = PrivateAttr(default=None)
+
+    @property
+    def client(self) -> MilvusClient:
+        """The Milvus client."""
+        return self._client
+
+    def add_document[D: Union[Dict[str, Any] | MilvusData]](self, collection_name: str, data: D | List[D]) -> Self:
+        """Adds a document to the specified collection.
+
+        Args:
+            collection_name (str): The name of the collection.
+            data (dict): The data to be added to the collection.
+        """
+        if isinstance(data, MilvusData):
+            data = data.prepare_insertion()
+        if isinstance(data, list):
+            data = [d.prepare_insertion() if isinstance(d, MilvusData) else d for d in data]
+        self.client.insert(collection_name, data)
+        return self

{fabricatio-0.2.3.dev0 → fabricatio-0.2.3.dev1}/python/fabricatio/config.py

@@ -11,6 +11,7 @@ from pydantic import (
     FilePath,
     HttpUrl,
     NonNegativeFloat,
+    PositiveFloat,
     PositiveInt,
     SecretStr,
 )
@@ -207,6 +208,19 @@ class ToolBoxConfig(BaseModel):
     """The name of the module containing the data."""
 
 
+class RagConfig(BaseModel):
+    """RAG configuration class."""
+
+    model_config = ConfigDict(use_attribute_docstrings=True)
+
+    milvus_uri: HttpUrl = Field(default=HttpUrl("http://localhost:19530"))
+    """The URI of the Milvus server."""
+    milvus_timeout: Optional[PositiveFloat] = Field(default=None)
+    """The timeout of the Milvus server."""
+    milvus_token: Optional[SecretStr] = Field(default=None)
+    """The token of the Milvus server."""
+
+
 class Settings(BaseSettings):
     """Application settings class.
 
@@ -250,6 +264,9 @@ class Settings(BaseSettings):
     toolbox: ToolBoxConfig = Field(default_factory=ToolBoxConfig)
     """Toolbox Configuration"""
 
+    rag: RagConfig = Field(default_factory=RagConfig)
+    """RAG Configuration"""
+
     @classmethod
     def settings_customise_sources(
         cls,

{fabricatio-0.2.3.dev0 → fabricatio-0.2.3.dev1}/python/fabricatio/models/kwargs_types.py

@@ -5,6 +5,15 @@ from typing import List, NotRequired, TypedDict
 from pydantic import NonNegativeFloat, NonNegativeInt, PositiveInt
 
 
+class EmbeddingKwargs(TypedDict):
+    """A type representing the keyword arguments for the embedding method."""
+
+    model: NotRequired[str]
+    dimensions: NotRequired[int]
+    timeout: NotRequired[PositiveInt]
+    caching: NotRequired[bool]
+
+
 class LLMKwargs(TypedDict):
     """A type representing the keyword arguments for the LLM (Large Language Model) usage."""
 

{fabricatio-0.2.3.dev0 → fabricatio-0.2.3.dev1}/python/fabricatio/models/usages.py

@@ -10,14 +10,15 @@ from fabricatio._rust_instances import template_manager
 from fabricatio.config import configs
 from fabricatio.journal import logger
 from fabricatio.models.generic import Base, WithBriefing
-from fabricatio.models.kwargs_types import ChooseKwargs, GenerateKwargs, LLMKwargs
+from fabricatio.models.kwargs_types import ChooseKwargs, EmbeddingKwargs, GenerateKwargs, LLMKwargs
 from fabricatio.models.task import Task
 from fabricatio.models.tool import Tool, ToolBox
-from fabricatio.models.utils import Messages
+from fabricatio.models.utils import Messages, MilvusData
 from fabricatio.parser import JsonCapture
 from litellm import stream_chunk_builder
 from litellm.types.utils import (
     Choices,
+    EmbeddingResponse,
     ModelResponse,
     StreamingChoices,
 )
@@ -61,6 +62,97 @@ class LLMUsage(Base):
     llm_max_tokens: Optional[PositiveInt] = None
     """The maximum number of tokens to generate."""
 
+    async def aembedding(
+        self,
+        input_text: List[str],
+        model: Optional[str] = None,
+        dimensions: Optional[int] = None,
+        timeout: Optional[PositiveInt] = None,
+        caching: Optional[bool] = False,
+    ) -> EmbeddingResponse:
+        """Asynchronously generates embeddings for the given input text.
+
+        Args:
+            input_text (List[str]): A list of strings to generate embeddings for.
+            model (Optional[str]): The model to use for embedding. Defaults to the instance's `llm_model` or the global configuration.
+            dimensions (Optional[int]): The dimensions of the embedding. Defaults to None.
+            timeout (Optional[PositiveInt]): The timeout for the embedding request. Defaults to the instance's `llm_timeout` or the global configuration.
+            caching (Optional[bool]): Whether to cache the embedding result. Defaults to False.
+
+
+        Returns:
+            EmbeddingResponse: The response containing the embeddings.
+        """
+        return await litellm.aembedding(
+            input=input_text,
+            caching=caching,
+            dimensions=dimensions,
+            model=model or self.llm_model or configs.llm.model,
+            timeout=timeout or self.llm_timeout or configs.llm.timeout,
+            api_key=self.llm_api_key.get_secret_value() if self.llm_api_key else configs.llm.api_key.get_secret_value(),
+            api_base=self.llm_api_endpoint.unicode_string().rstrip(
+                "/"
+            )  # seems embedding function takes no base_url end with a slash
+            if self.llm_api_endpoint
+            else configs.llm.api_endpoint.unicode_string().rstrip("/"),
+        )
+
+    @overload
+    async def vectorize(self, input_text: List[str], **kwargs: Unpack[EmbeddingKwargs]) -> List[List[float]]: ...
+    @overload
+    async def vectorize(self, input_text: str, **kwargs: Unpack[EmbeddingKwargs]) -> List[float]: ...
+
+    async def vectorize(
+        self, input_text: List[str] | str, **kwargs: Unpack[EmbeddingKwargs]
+    ) -> List[List[float]] | List[float]:
+        """Asynchronously generates vector embeddings for the given input text.
+
+        Args:
+            input_text (List[str] | str): A string or list of strings to generate embeddings for.
+            **kwargs (Unpack[EmbeddingKwargs]): Additional keyword arguments for embedding.
+
+        Returns:
+            List[List[float]] | List[float]: The generated embeddings.
+        """
+        if isinstance(input_text, str):
+            return (await self.aembedding([input_text], **kwargs)).data[0].get("embedding")
+
+        return [o.get("embedding") for o in (await self.aembedding(input_text, **kwargs)).data]
+
+    @overload
+    async def pack(
+        self, input_text: List[str], subject: Optional[str] = None, **kwargs: Unpack[EmbeddingKwargs]
+    ) -> List[MilvusData]: ...
+    @overload
+    async def pack(
+        self, input_text: str, subject: Optional[str] = None, **kwargs: Unpack[EmbeddingKwargs]
+    ) -> MilvusData: ...
+
+    async def pack(
+        self, input_text: List[str] | str, subject: Optional[str] = None, **kwargs: Unpack[EmbeddingKwargs]
+    ) -> List[MilvusData] | MilvusData:
+        """Asynchronously generates MilvusData objects for the given input text.
+
+        Args:
+            input_text (List[str] | str): A string or list of strings to generate embeddings for.
+            subject (Optional[str]): The subject of the input text. Defaults to None.
+            **kwargs (Unpack[EmbeddingKwargs]): Additional keyword arguments for embedding.
+
+        Returns:
+            List[MilvusData] | MilvusData: The generated MilvusData objects.
+        """
+        if isinstance(input_text, str):
+            return MilvusData(vector=await self.vectorize(input_text, **kwargs), text=input_text, subject=subject)
+        vecs = await self.vectorize(input_text, **kwargs)
+        return [
+            MilvusData(
+                vector=vec,
+                text=text,
+                subject=subject,
+            )
+            for text, vec in zip(input_text, vecs, strict=True)
+        ]
+
     async def aquery(
         self,
         messages: List[Dict[str, str]],

{fabricatio-0.2.3.dev0 → fabricatio-0.2.3.dev1}/python/fabricatio/models/utils.py

@@ -1,6 +1,6 @@
 """A module containing utility classes for the models."""
 
-from typing import Dict, List, Literal, Self
+from typing import Any, Dict, List, Literal, Optional, Self
 
 from pydantic import BaseModel, ConfigDict, Field
 
@@ -76,3 +76,52 @@ class Messages(list):
             list[dict]: A list of dictionaries representing the messages.
         """
         return [message.model_dump() for message in self]
+
+
+class MilvusData(BaseModel):
+    """A class representing data stored in Milvus."""
+
+    model_config = ConfigDict(use_attribute_docstrings=True)
+    id: Optional[int] = Field(default=None)
+    """The identifier of the data."""
+
+    vector: List[float]
+    """The vector representation of the data."""
+
+    text: str
+    """The text representation of the data."""
+
+    subject: Optional[str] = Field(default=None)
+    """A subject label that we use to demo metadata filtering later."""
+
+    def prepare_insertion(self) -> Dict[str, Any]:
+        """Prepares the data for insertion into Milvus.
+
+        Returns:
+            dict: A dictionary containing the data to be inserted into Milvus.
+        """
+        return self.model_dump(exclude_none=True)
+
+    def update_subject(self, new_subject: str) -> Self:
+        """Updates the subject label of the data.
+
+        Args:
+            new_subject (str): The new subject label.
+
+        Returns:
+            Self: The updated instance of MilvusData.
+        """
+        self.subject = new_subject
+        return self
+
+    def update_id(self, new_id: int) -> Self:
+        """Updates the identifier of the data.
+
+        Args:
+            new_id (int): The new identifier.
+
+        Returns:
+            Self: The updated instance of MilvusData.
+        """
+        self.id = new_id
+        return self