langchain 0.2.9__tar.gz → 0.2.11__tar.gz

{langchain-0.2.9 → langchain-0.2.11}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: langchain
-Version: 0.2.9
+Version: 0.2.11
 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 (>=0.2.23,<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"
@@ -24,6 +24,8 @@ Requires-Dist: pydantic (>=1,<3)
 Requires-Dist: requests (>=2,<3)
 Requires-Dist: tenacity (>=8.1.0,<9.0.0,!=8.4.0)
 Project-URL: Repository, https://github.com/langchain-ai/langchain
+Project-URL: Release Notes, https://github.com/langchain-ai/langchain/releases?q=tag%3A%22langchain%3D%3D0%22&expanded=true
+Project-URL: Source Code, https://github.com/langchain-ai/langchain/tree/master/libs/langchain
 Description-Content-Type: text/markdown
 
 # 🦜️🔗 LangChain

{langchain-0.2.9 → langchain-0.2.11}/langchain/agents/agent.py

@@ -1146,30 +1146,6 @@ class AgentExecutor(Chain):
                 )
         return values
 
-    @root_validator(pre=False, skip_on_failure=True)
-    def validate_return_direct_tool(cls, values: Dict) -> Dict:
-        """Validate that tools are compatible with agent.
-
-        Args:
-            values: Values to validate.
-
-        Returns:
-            Dict: Validated values.
-
-        Raises:
-            ValueError: If tools that have `return_direct=True` are not allowed.
-        """
-        agent = values["agent"]
-        tools = values["tools"]
-        if isinstance(agent, BaseMultiActionAgent):
-            for tool in tools:
-                if tool.return_direct:
-                    raise ValueError(
-                        "Tools that have `return_direct=True` are not allowed "
-                        "in multi-action agents"
-                    )
-        return values
-
     @root_validator(pre=True)
     def validate_runnable_agent(cls, values: Dict) -> Dict:
         """Convert runnable to agent if passed in.

{langchain-0.2.9 → langchain-0.2.11}/langchain/agents/react/agent.py

@@ -71,7 +71,7 @@ def create_react_agent(
                     "input": "what's my name?",
                     # Notice that chat_history is a string
                     # since this prompt is aimed at LLMs, not chat models
-                    "chat_history": "Human: My name is Bob\nAI: Hello Bob!",
+                    "chat_history": "Human: My name is Bob\\nAI: Hello Bob!",
                 }
             )
 

{langchain-0.2.9 → langchain-0.2.11}/langchain/chains/conversational_retrieval/base.py

@@ -42,8 +42,11 @@ def _get_chat_history(chat_history: List[CHAT_TURN_TYPE]) -> str:
     buffer = ""
     for dialogue_turn in chat_history:
         if isinstance(dialogue_turn, BaseMessage):
-            role_prefix = _ROLE_MAP.get(dialogue_turn.type, f"{dialogue_turn.type}: ")
-            buffer += f"\n{role_prefix}{dialogue_turn.content}"
+            if len(dialogue_turn.content) > 0:
+                role_prefix = _ROLE_MAP.get(
+                    dialogue_turn.type, f"{dialogue_turn.type}: "
+                )
+                buffer += f"\n{role_prefix}{dialogue_turn.content}"
         elif isinstance(dialogue_turn, tuple):
             human = "Human: " + dialogue_turn[0]
             ai = "Assistant: " + dialogue_turn[1]

{langchain-0.2.9 → langchain-0.2.11}/langchain/chains/openai_functions/qa_with_structure.py

@@ -1,4 +1,4 @@
-from typing import Any, List, Optional, Type, Union
+from typing import Any, List, Optional, Type, Union, cast
 
 from langchain_core.language_models import BaseLanguageModel
 from langchain_core.messages import HumanMessage, SystemMessage
@@ -10,6 +10,7 @@ from langchain_core.output_parsers.openai_functions import (
 from langchain_core.prompts import PromptTemplate
 from langchain_core.prompts.chat import ChatPromptTemplate, HumanMessagePromptTemplate
 from langchain_core.pydantic_v1 import BaseModel, Field
+from langchain_core.utils.pydantic import is_basemodel_subclass
 
 from langchain.chains.llm import LLMChain
 from langchain.chains.openai_functions.utils import get_llm_kwargs
@@ -45,7 +46,7 @@ def create_qa_with_structure_chain(
 
     """
     if output_parser == "pydantic":
-        if not (isinstance(schema, type) and issubclass(schema, BaseModel)):
+        if not (isinstance(schema, type) and is_basemodel_subclass(schema)):
             raise ValueError(
                 "Must provide a pydantic class for schema when output_parser is "
                 "'pydantic'."
@@ -60,10 +61,10 @@ def create_qa_with_structure_chain(
             f"Got unexpected output_parser: {output_parser}. "
             f"Should be one of `pydantic` or `base`."
         )
-    if isinstance(schema, type) and issubclass(schema, BaseModel):
-        schema_dict = schema.schema()
+    if isinstance(schema, type) and is_basemodel_subclass(schema):
+        schema_dict = cast(dict, schema.schema())
     else:
-        schema_dict = schema
+        schema_dict = cast(dict, schema)
     function = {
         "name": schema_dict["title"],
         "description": schema_dict["description"],

{langchain-0.2.9 → langchain-0.2.11}/langchain/chains/structured_output/base.py

@@ -24,6 +24,7 @@ from langchain_core.utils.function_calling import (
     convert_to_openai_function,
     convert_to_openai_tool,
 )
+from langchain_core.utils.pydantic import is_basemodel_subclass
 
 
 @deprecated(
@@ -465,7 +466,7 @@ def _get_openai_tool_output_parser(
     *,
     first_tool_only: bool = False,
 ) -> Union[BaseOutputParser, BaseGenerationOutputParser]:
-    if isinstance(tool, type) and issubclass(tool, BaseModel):
+    if isinstance(tool, type) and is_basemodel_subclass(tool):
         output_parser: Union[BaseOutputParser, BaseGenerationOutputParser] = (
             PydanticToolsParser(tools=[tool], first_tool_only=first_tool_only)
         )
@@ -493,7 +494,7 @@ def get_openai_output_parser(
             not a Pydantic class, then the output parser will automatically extract
             only the function arguments and not the function name.
     """
-    if isinstance(functions[0], type) and issubclass(functions[0], BaseModel):
+    if isinstance(functions[0], type) and is_basemodel_subclass(functions[0]):
         if len(functions) > 1:
             pydantic_schema: Union[Dict, Type[BaseModel]] = {
                 convert_to_openai_function(fn)["name"]: fn for fn in functions
@@ -516,7 +517,7 @@ def _create_openai_json_runnable(
     output_parser: Optional[Union[BaseOutputParser, BaseGenerationOutputParser]] = None,
 ) -> Runnable:
     """"""
-    if isinstance(output_schema, type) and issubclass(output_schema, BaseModel):
+    if isinstance(output_schema, type) and is_basemodel_subclass(output_schema):
         output_parser = output_parser or PydanticOutputParser(
             pydantic_object=output_schema,  # type: ignore
         )

{langchain-0.2.9 → langchain-0.2.11}/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-0.2.11}/langchain/memory/summary.py

@@ -34,6 +34,18 @@ class SummarizerMixin(BaseModel):
         chain = LLMChain(llm=self.llm, prompt=self.prompt)
         return chain.predict(summary=existing_summary, new_lines=new_lines)
 
+    async def apredict_new_summary(
+        self, messages: List[BaseMessage], existing_summary: str
+    ) -> str:
+        new_lines = get_buffer_string(
+            messages,
+            human_prefix=self.human_prefix,
+            ai_prefix=self.ai_prefix,
+        )
+
+        chain = LLMChain(llm=self.llm, prompt=self.prompt)
+        return await chain.apredict(summary=existing_summary, new_lines=new_lines)
+
 
 class ConversationSummaryMemory(BaseChatMemory, SummarizerMixin):
     """Conversation summarizer to chat memory."""

{langchain-0.2.9 → langchain-0.2.11}/langchain/memory/summary_buffer.py

@@ -19,6 +19,11 @@ class ConversationSummaryBufferMemory(BaseChatMemory, SummarizerMixin):
         """String buffer of memory."""
         return self.load_memory_variables({})[self.memory_key]
 
+    async def abuffer(self) -> Union[str, List[BaseMessage]]:
+        """Async memory buffer."""
+        memory_variables = await self.aload_memory_variables({})
+        return memory_variables[self.memory_key]
+
     @property
     def memory_variables(self) -> List[str]:
         """Will always return list of memory variables.
@@ -43,6 +48,22 @@ class ConversationSummaryBufferMemory(BaseChatMemory, SummarizerMixin):
             )
         return {self.memory_key: final_buffer}
 
+    async def aload_memory_variables(self, inputs: Dict[str, Any]) -> Dict[str, Any]:
+        """Asynchronously return key-value pairs given the text input to the chain."""
+        buffer = await self.chat_memory.aget_messages()
+        if self.moving_summary_buffer != "":
+            first_messages: List[BaseMessage] = [
+                self.summary_message_cls(content=self.moving_summary_buffer)
+            ]
+            buffer = first_messages + buffer
+        if self.return_messages:
+            final_buffer: Any = buffer
+        else:
+            final_buffer = get_buffer_string(
+                buffer, human_prefix=self.human_prefix, ai_prefix=self.ai_prefix
+            )
+        return {self.memory_key: final_buffer}
+
     @root_validator()
     def validate_prompt_input_variables(cls, values: Dict) -> Dict:
         """Validate that prompt input variables are consistent."""
@@ -60,6 +81,13 @@ class ConversationSummaryBufferMemory(BaseChatMemory, SummarizerMixin):
         super().save_context(inputs, outputs)
         self.prune()
 
+    async def asave_context(
+        self, inputs: Dict[str, Any], outputs: Dict[str, str]
+    ) -> None:
+        """Asynchronously save context from this conversation to buffer."""
+        await super().asave_context(inputs, outputs)
+        await self.aprune()
+
     def prune(self) -> None:
         """Prune buffer if it exceeds max token limit"""
         buffer = self.chat_memory.messages
@@ -73,7 +101,25 @@ class ConversationSummaryBufferMemory(BaseChatMemory, SummarizerMixin):
                 pruned_memory, self.moving_summary_buffer
             )
 
+    async def aprune(self) -> None:
+        """Asynchronously prune buffer if it exceeds max token limit"""
+        buffer = self.chat_memory.messages
+        curr_buffer_length = self.llm.get_num_tokens_from_messages(buffer)
+        if curr_buffer_length > self.max_token_limit:
+            pruned_memory = []
+            while curr_buffer_length > self.max_token_limit:
+                pruned_memory.append(buffer.pop(0))
+                curr_buffer_length = self.llm.get_num_tokens_from_messages(buffer)
+            self.moving_summary_buffer = await self.apredict_new_summary(
+                pruned_memory, self.moving_summary_buffer
+            )
+
     def clear(self) -> None:
         """Clear memory contents."""
         super().clear()
         self.moving_summary_buffer = ""
+
+    async def aclear(self) -> None:
+        """Asynchronously clear memory contents."""
+        await super().aclear()
+        self.moving_summary_buffer = ""

{langchain-0.2.9 → langchain-0.2.11}/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.
@@ -67,7 +67,7 @@ class ConversationVectorStoreTokenBufferMemory(ConversationTokenBufferMemory):
         from langchain.memory.token_buffer_vectorstore_memory import (
                 ConversationVectorStoreTokenBufferMemory
         )
-        from langchain_community.vectorstores import Chroma
+        from langchain_chroma import Chroma
         from langchain_community.embeddings import HuggingFaceInstructEmbeddings
         from langchain_openai import OpenAI
 

{langchain-0.2.9 → langchain-0.2.11}/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-0.2.11}/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-0.2.11}/langchain/retrievers/contextual_compression.py

@@ -5,7 +5,7 @@ from langchain_core.callbacks import (
     CallbackManagerForRetrieverRun,
 )
 from langchain_core.documents import Document
-from langchain_core.retrievers import BaseRetriever
+from langchain_core.retrievers import BaseRetriever, RetrieverLike
 
 from langchain.retrievers.document_compressors.base import (
     BaseDocumentCompressor,
@@ -18,7 +18,7 @@ class ContextualCompressionRetriever(BaseRetriever):
     base_compressor: BaseDocumentCompressor
     """Compressor for compressing retrieved documents."""
 
-    base_retriever: BaseRetriever
+    base_retriever: RetrieverLike
     """Base Retriever to use for getting relevant documents."""
 
     class Config:

{langchain-0.2.9 → langchain-0.2.11}/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-0.2.11/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-0.2.11}/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