langchain 0.2.9__tar.gz → 0.2.10__tar.gz

{langchain-0.2.9 → langchain-0.2.10}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: langchain
-Version: 0.2.9
+Version: 0.2.10
 Summary: Building applications with LLMs through composability
 Home-page: https://github.com/langchain-ai/langchain
 License: MIT
@@ -15,7 +15,7 @@ Requires-Dist: PyYAML (>=5.3)
 Requires-Dist: SQLAlchemy (>=1.4,<3)
 Requires-Dist: aiohttp (>=3.8.3,<4.0.0)
 Requires-Dist: async-timeout (>=4.0.0,<5.0.0) ; python_version < "3.11"
-Requires-Dist: langchain-core (>=0.2.20,<0.3.0)
+Requires-Dist: langchain-core (>=0.2.22,<0.3.0)
 Requires-Dist: langchain-text-splitters (>=0.2.0,<0.3.0)
 Requires-Dist: langsmith (>=0.1.17,<0.2.0)
 Requires-Dist: numpy (>=1,<2) ; python_version < "3.12"

{langchain-0.2.9 → langchain-0.2.10}/langchain/evaluation/loading.py

@@ -146,7 +146,7 @@ def load_evaluator(
                     )
 
             llm = llm or ChatOpenAI(  # type: ignore[call-arg]
-                model="gpt-4", model_kwargs={"seed": 42}, temperature=0
+                model="gpt-4", seed=42, temperature=0
             )
         except Exception as e:
             raise ValueError(

{langchain-0.2.9 → langchain-0.2.10}/langchain/memory/vectorstore_token_buffer_memory.py

@@ -1,7 +1,7 @@
 """
 Class for a conversation memory buffer with older messages stored in a vectorstore .
 
-This implementats a conversation memory in which the messages are stored in a memory
+This implements a conversation memory in which the messages are stored in a memory
 buffer up to a specified token limit. When the limit is exceeded, older messages are
 saved to a vectorstore backing database. The vectorstore can be made persistent across
 sessions.

{langchain-0.2.9 → langchain-0.2.10}/langchain/output_parsers/fix.py

@@ -7,12 +7,19 @@ from langchain_core.language_models import BaseLanguageModel
 from langchain_core.output_parsers import BaseOutputParser
 from langchain_core.prompts import BasePromptTemplate
 from langchain_core.runnables import RunnableSerializable
+from typing_extensions import TypedDict
 
 from langchain.output_parsers.prompts import NAIVE_FIX_PROMPT
 
 T = TypeVar("T")
 
 
+class OutputFixingParserRetryChainInput(TypedDict, total=False):
+    instructions: str
+    completion: str
+    error: str
+
+
 class OutputFixingParser(BaseOutputParser[T]):
     """Wrap a parser and try to fix parsing errors."""
 
@@ -23,7 +30,9 @@ class OutputFixingParser(BaseOutputParser[T]):
     parser: BaseOutputParser[T]
     """The parser to use to parse the output."""
     # Should be an LLMChain but we want to avoid top-level imports from langchain.chains
-    retry_chain: Union[RunnableSerializable, Any]
+    retry_chain: Union[
+        RunnableSerializable[OutputFixingParserRetryChainInput, str], Any
+    ]
     """The RunnableSerializable to use to retry the completion (Legacy: LLMChain)."""
     max_retries: int = 1
     """The maximum number of times to retry the parse."""
@@ -73,16 +82,16 @@ class OutputFixingParser(BaseOutputParser[T]):
                         try:
                             completion = self.retry_chain.invoke(
                                 dict(
-                                    instructions=self.parser.get_format_instructions(),  # noqa: E501
-                                    input=completion,
+                                    instructions=self.parser.get_format_instructions(),
+                                    completion=completion,
                                     error=repr(e),
                                 )
                             )
                         except (NotImplementedError, AttributeError):
-                            # Case: self.parser does not have get_format_instructions  # noqa: E501
+                            # Case: self.parser does not have get_format_instructions
                             completion = self.retry_chain.invoke(
                                 dict(
-                                    input=completion,
+                                    completion=completion,
                                     error=repr(e),
                                 )
                             )
@@ -102,7 +111,7 @@ class OutputFixingParser(BaseOutputParser[T]):
                     retries += 1
                     if self.legacy and hasattr(self.retry_chain, "arun"):
                         completion = await self.retry_chain.arun(
-                            instructions=self.parser.get_format_instructions(),  # noqa: E501
+                            instructions=self.parser.get_format_instructions(),
                             completion=completion,
                             error=repr(e),
                         )
@@ -110,16 +119,16 @@ class OutputFixingParser(BaseOutputParser[T]):
                         try:
                             completion = await self.retry_chain.ainvoke(
                                 dict(
-                                    instructions=self.parser.get_format_instructions(),  # noqa: E501
-                                    input=completion,
+                                    instructions=self.parser.get_format_instructions(),
+                                    completion=completion,
                                     error=repr(e),
                                 )
                             )
                         except (NotImplementedError, AttributeError):
-                            # Case: self.parser does not have get_format_instructions  # noqa: E501
+                            # Case: self.parser does not have get_format_instructions
                             completion = await self.retry_chain.ainvoke(
                                 dict(
-                                    input=completion,
+                                    completion=completion,
                                     error=repr(e),
                                 )
                             )

{langchain-0.2.9 → langchain-0.2.10}/langchain/output_parsers/retry.py

@@ -8,6 +8,7 @@ from langchain_core.output_parsers import BaseOutputParser
 from langchain_core.prompt_values import PromptValue
 from langchain_core.prompts import BasePromptTemplate, PromptTemplate
 from langchain_core.runnables import RunnableSerializable
+from typing_extensions import TypedDict
 
 NAIVE_COMPLETION_RETRY = """Prompt:
 {prompt}
@@ -34,6 +35,17 @@ NAIVE_RETRY_WITH_ERROR_PROMPT = PromptTemplate.from_template(
 T = TypeVar("T")
 
 
+class RetryOutputParserRetryChainInput(TypedDict):
+    prompt: str
+    completion: str
+
+
+class RetryWithErrorOutputParserRetryChainInput(TypedDict):
+    prompt: str
+    completion: str
+    error: str
+
+
 class RetryOutputParser(BaseOutputParser[T]):
     """Wrap a parser and try to fix parsing errors.
 
@@ -44,7 +56,7 @@ class RetryOutputParser(BaseOutputParser[T]):
     parser: BaseOutputParser[T]
     """The parser to use to parse the output."""
     # Should be an LLMChain but we want to avoid top-level imports from langchain.chains
-    retry_chain: Union[RunnableSerializable, Any]
+    retry_chain: Union[RunnableSerializable[RetryOutputParserRetryChainInput, str], Any]
     """The RunnableSerializable to use to retry the completion (Legacy: LLMChain)."""
     max_retries: int = 1
     """The maximum number of times to retry the parse."""
@@ -97,13 +109,12 @@ class RetryOutputParser(BaseOutputParser[T]):
                         completion = self.retry_chain.run(
                             prompt=prompt_value.to_string(),
                             completion=completion,
-                            error=repr(e),
                         )
                     else:
                         completion = self.retry_chain.invoke(
                             dict(
                                 prompt=prompt_value.to_string(),
-                                input=completion,
+                                completion=completion,
                             )
                         )
 
@@ -139,7 +150,7 @@ class RetryOutputParser(BaseOutputParser[T]):
                         completion = await self.retry_chain.ainvoke(
                             dict(
                                 prompt=prompt_value.to_string(),
-                                input=completion,
+                                completion=completion,
                             )
                         )
 
@@ -174,8 +185,10 @@ class RetryWithErrorOutputParser(BaseOutputParser[T]):
 
     parser: BaseOutputParser[T]
     """The parser to use to parse the output."""
-    # Should be an LLMChain but we want to avoid top-level imports from langchain.chains  # noqa: E501
-    retry_chain: Union[RunnableSerializable, Any]
+    # Should be an LLMChain but we want to avoid top-level imports from langchain.chains
+    retry_chain: Union[
+        RunnableSerializable[RetryWithErrorOutputParserRetryChainInput, str], Any
+    ]
     """The RunnableSerializable to use to retry the completion (Legacy: LLMChain)."""
     max_retries: int = 1
     """The maximum number of times to retry the parse."""
@@ -204,7 +217,7 @@ class RetryWithErrorOutputParser(BaseOutputParser[T]):
         chain = prompt | llm
         return cls(parser=parser, retry_chain=chain, max_retries=max_retries)
 
-    def parse_with_prompt(self, completion: str, prompt_value: PromptValue) -> T:  # noqa: E501
+    def parse_with_prompt(self, completion: str, prompt_value: PromptValue) -> T:
         retries = 0
 
         while retries <= self.max_retries:
@@ -224,7 +237,7 @@ class RetryWithErrorOutputParser(BaseOutputParser[T]):
                     else:
                         completion = self.retry_chain.invoke(
                             dict(
-                                input=completion,
+                                completion=completion,
                                 prompt=prompt_value.to_string(),
                                 error=repr(e),
                             )
@@ -253,7 +266,7 @@ class RetryWithErrorOutputParser(BaseOutputParser[T]):
                         completion = await self.retry_chain.ainvoke(
                             dict(
                                 prompt=prompt_value.to_string(),
-                                input=completion,
+                                completion=completion,
                                 error=repr(e),
                             )
                         )

{langchain-0.2.9 → langchain-0.2.10}/langchain/retrievers/document_compressors/__init__.py

@@ -15,6 +15,9 @@ from langchain.retrievers.document_compressors.cross_encoder_rerank import (
 from langchain.retrievers.document_compressors.embeddings_filter import (
     EmbeddingsFilter,
 )
+from langchain.retrievers.document_compressors.listwise_rerank import (
+    LLMListwiseRerank,
+)
 
 _module_lookup = {
     "FlashrankRerank": "langchain_community.document_compressors.flashrank_rerank",
@@ -31,6 +34,7 @@ def __getattr__(name: str) -> Any:
 __all__ = [
     "DocumentCompressorPipeline",
     "EmbeddingsFilter",
+    "LLMListwiseRerank",
     "LLMChainExtractor",
     "LLMChainFilter",
     "CohereRerank",

langchain-0.2.10/langchain/retrievers/document_compressors/listwise_rerank.py

@@ -0,0 +1,137 @@
+"""Filter that uses an LLM to rerank documents listwise and select top-k."""
+
+from typing import Any, Dict, List, Optional, Sequence
+
+from langchain_core.callbacks import Callbacks
+from langchain_core.documents import BaseDocumentCompressor, Document
+from langchain_core.language_models import BaseLanguageModel
+from langchain_core.prompts import BasePromptTemplate, ChatPromptTemplate
+from langchain_core.pydantic_v1 import BaseModel, Field
+from langchain_core.runnables import Runnable, RunnableLambda, RunnablePassthrough
+
+_default_system_tmpl = """{context}
+
+Sort the Documents by their relevance to the Query."""
+_DEFAULT_PROMPT = ChatPromptTemplate.from_messages(
+    [("system", _default_system_tmpl), ("human", "{query}")],
+)
+
+
+def _get_prompt_input(input_: dict) -> Dict[str, Any]:
+    """Return the compression chain input."""
+    documents = input_["documents"]
+    context = ""
+    for index, doc in enumerate(documents):
+        context += f"Document ID: {index}\n```{doc.page_content}```\n\n"
+    context += f"Documents = [Document ID: 0, ..., Document ID: {len(documents) - 1}]"
+    return {"query": input_["query"], "context": context}
+
+
+def _parse_ranking(results: dict) -> List[Document]:
+    ranking = results["ranking"]
+    docs = results["documents"]
+    return [docs[i] for i in ranking.ranked_document_ids]
+
+
+class LLMListwiseRerank(BaseDocumentCompressor):
+    """Document compressor that uses `Zero-Shot Listwise Document Reranking`.
+
+    Adapted from: https://arxiv.org/pdf/2305.02156.pdf
+
+    ``LLMListwiseRerank`` uses a language model to rerank a list of documents based on
+    their relevance to a query.
+
+    **NOTE**: requires that underlying model implement ``with_structured_output``.
+
+    Example usage:
+        .. code-block:: python
+
+            from langchain.retrievers.document_compressors.listwise_rerank import (
+                LLMListwiseRerank,
+            )
+            from langchain_core.documents import Document
+            from langchain_openai import ChatOpenAI
+
+            documents = [
+                Document("Sally is my friend from school"),
+                Document("Steve is my friend from home"),
+                Document("I didn't always like yogurt"),
+                Document("I wonder why it's called football"),
+                Document("Where's waldo"),
+            ]
+
+            reranker = LLMListwiseRerank.from_llm(
+                llm=ChatOpenAI(model="gpt-3.5-turbo"), top_n=3
+            )
+            compressed_docs = reranker.compress_documents(documents, "Who is steve")
+            assert len(compressed_docs) == 3
+            assert "Steve" in compressed_docs[0].page_content
+    """
+
+    reranker: Runnable[Dict, List[Document]]
+    """LLM-based reranker to use for filtering documents. Expected to take in a dict 
+        with 'documents: Sequence[Document]' and 'query: str' keys and output a 
+        List[Document]."""
+
+    top_n: int = 3
+    """Number of documents to return."""
+
+    class Config:
+        arbitrary_types_allowed = True
+
+    def compress_documents(
+        self,
+        documents: Sequence[Document],
+        query: str,
+        callbacks: Optional[Callbacks] = None,
+    ) -> Sequence[Document]:
+        """Filter down documents based on their relevance to the query."""
+        results = self.reranker.invoke(
+            {"documents": documents, "query": query}, config={"callbacks": callbacks}
+        )
+        return results[: self.top_n]
+
+    @classmethod
+    def from_llm(
+        cls,
+        llm: BaseLanguageModel,
+        *,
+        prompt: Optional[BasePromptTemplate] = None,
+        **kwargs: Any,
+    ) -> "LLMListwiseRerank":
+        """Create a LLMListwiseRerank document compressor from a language model.
+
+        Args:
+            llm: The language model to use for filtering. **Must implement
+                BaseLanguageModel.with_structured_output().**
+            prompt: The prompt to use for the filter.
+            **kwargs: Additional arguments to pass to the constructor.
+
+        Returns:
+            A LLMListwiseRerank document compressor that uses the given language model.
+        """
+
+        if llm.with_structured_output == BaseLanguageModel.with_structured_output:
+            raise ValueError(
+                f"llm of type {type(llm)} does not implement `with_structured_output`."
+            )
+
+        class RankDocuments(BaseModel):
+            """Rank the documents by their relevance to the user question.
+            Rank from most to least relevant."""
+
+            ranked_document_ids: List[int] = Field(
+                ...,
+                description=(
+                    "The integer IDs of the documents, sorted from most to least "
+                    "relevant to the user question."
+                ),
+            )
+
+        _prompt = prompt if prompt is not None else _DEFAULT_PROMPT
+        reranker = RunnablePassthrough.assign(
+            ranking=RunnableLambda(_get_prompt_input)
+            | _prompt
+            | llm.with_structured_output(RankDocuments)
+        ) | RunnableLambda(_parse_ranking)
+        return cls(reranker=reranker, **kwargs)

{langchain-0.2.9 → langchain-0.2.10}/langchain/retrievers/multi_query.py

@@ -24,7 +24,7 @@ class LineListOutputParser(BaseOutputParser[List[str]]):
 
     def parse(self, text: str) -> List[str]:
         lines = text.strip().split("\n")
-        return lines
+        return list(filter(None, lines))  # Remove empty lines
 
 
 # Default prompt

{langchain-0.2.9 → langchain-0.2.10}/langchain/retrievers/parent_document_retriever.py

@@ -1,5 +1,5 @@
 import uuid
-from typing import Any, List, Optional, Sequence
+from typing import Any, List, Optional, Sequence, Tuple
 
 from langchain_core.documents import Document
 from langchain_text_splitters import TextSplitter
@@ -69,27 +69,12 @@ class ParentDocumentRetriever(MultiVectorRetriever):
         metadata.
     """
 
-    def add_documents(
+    def _split_docs_for_adding(
         self,
         documents: List[Document],
         ids: Optional[List[str]] = None,
         add_to_docstore: bool = True,
-        **kwargs: Any,
-    ) -> None:
-        """Adds documents to the docstore and vectorstores.
-
-        Args:
-            documents: List of documents to add
-            ids: Optional list of ids for documents. If provided should be the same
-                length as the list of documents. Can be provided if parent documents
-                are already in the document store and you don't want to re-add
-                to the docstore. If not provided, random UUIDs will be used as
-                ids.
-            add_to_docstore: Boolean of whether to add documents to docstore.
-                This can be false if and only if `ids` are provided. You may want
-                to set this to False if the documents are already in the docstore
-                and you don't want to re-add them.
-        """
+    ) -> Tuple[List[Document], List[Tuple[str, Document]]]:
         if self.parent_splitter is not None:
             documents = self.parent_splitter.split_documents(documents)
         if ids is None:
@@ -120,6 +105,43 @@ class ParentDocumentRetriever(MultiVectorRetriever):
                 _doc.metadata[self.id_key] = _id
             docs.extend(sub_docs)
             full_docs.append((_id, doc))
+
+        return docs, full_docs
+
+    def add_documents(
+        self,
+        documents: List[Document],
+        ids: Optional[List[str]] = None,
+        add_to_docstore: bool = True,
+        **kwargs: Any,
+    ) -> None:
+        """Adds documents to the docstore and vectorstores.
+
+        Args:
+            documents: List of documents to add
+            ids: Optional list of ids for documents. If provided should be the same
+                length as the list of documents. Can be provided if parent documents
+                are already in the document store and you don't want to re-add
+                to the docstore. If not provided, random UUIDs will be used as
+                ids.
+            add_to_docstore: Boolean of whether to add documents to docstore.
+                This can be false if and only if `ids` are provided. You may want
+                to set this to False if the documents are already in the docstore
+                and you don't want to re-add them.
+        """
+        docs, full_docs = self._split_docs_for_adding(documents, ids, add_to_docstore)
         self.vectorstore.add_documents(docs, **kwargs)
         if add_to_docstore:
             self.docstore.mset(full_docs)
+
+    async def aadd_documents(
+        self,
+        documents: List[Document],
+        ids: Optional[List[str]] = None,
+        add_to_docstore: bool = True,
+        **kwargs: Any,
+    ) -> None:
+        docs, full_docs = self._split_docs_for_adding(documents, ids, add_to_docstore)
+        await self.vectorstore.aadd_documents(docs, **kwargs)
+        if add_to_docstore:
+            await self.docstore.amset(full_docs)

{langchain-0.2.9 → langchain-0.2.10}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "poetry.core.masonry.api"
 
 [tool.poetry]
 name = "langchain"
-version = "0.2.9"
+version = "0.2.10"
 description = "Building applications with LLMs through composability"
 authors = []
 license = "MIT"
@@ -29,7 +29,7 @@ langchain-server = "langchain.server:main"
 
 [tool.poetry.dependencies]
 python = ">=3.8.1,<4.0"
-langchain-core = "^0.2.20"
+langchain-core = "^0.2.22"
 langchain-text-splitters = "^0.2.0"
 langsmith = "^0.1.17"
 pydantic = ">=1,<3"