Add dense proposals (#14719)

Indexing strategy based on decomposing candidate propositions while
indexing.

templates/rag-chroma-dense-retrieval/.gitignore

@@ -0,0 +1,3 @@
+docs/img_*.jpg
+chroma_db_proposals
+multi_vector_retriever_metadata

templates/rag-chroma-dense-retrieval/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2023 LangChain, Inc.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

templates/rag-chroma-dense-retrieval/README.md

@@ -0,0 +1,81 @@
+# rag-chroma-dense-retrieval
+
+This template demonstrates the multi-vector indexing strategy proposed by Chen, et. al.'s [Dense X Retrieval: What Retrieval Granularity Should We Use?](https://arxiv.org/abs/2312.06648). The prompt, which you can [try out on the hub](https://smith.langchain.com/hub/wfh/proposal-indexing), directs an LLM to generate de-contextualized "propositions" which can be vectorized to increase the retrieval accuracy. You can see the full definition in `proposal_chain.py`.
+
+![Retriever Diagram](./_images/retriever_diagram.png)
+
+## Storage
+
+For this demo, we index a simple academic paper using the RecursiveUrlLoader, and store all retriever information locally (using chroma and a bytestore stored on the local filesystem). You can modify the storage layer in `storage.py`.
+
+## Environment Setup
+
+Set the `OPENAI_API_KEY` environment variable to access `gpt-3.5` and the OpenAI Embeddings classes.
+
+## Indexing
+
+Create the index by running the following:
+
+```python
+poetry install
+poetry run python rag_chroma_dense_retrieval/ingest.py
+```
+
+## Usage
+
+To use this package, you should first have the LangChain CLI installed:
+
+```shell
+pip install -U langchain-cli
+```
+
+To create a new LangChain project and install this as the only package, you can do:
+
+```shell
+langchain app new my-app --package rag-chroma-dense-retrieval
+```
+
+If you want to add this to an existing project, you can just run:
+
+```shell
+langchain app add rag-chroma-dense-retrieval
+```
+
+And add the following code to your `server.py` file:
+
+```python
+from rag_chroma_dense_retrieval import chain
+
+add_routes(app, chain, path="/rag-chroma-dense-retrieval")
+```
+
+(Optional) Let's now configure LangSmith.
+LangSmith will help us trace, monitor and debug LangChain applications.
+LangSmith is currently in private beta, you can sign up [here](https://smith.langchain.com/).
+If you don't have access, you can skip this section
+
+```shell
+export LANGCHAIN_TRACING_V2=true
+export LANGCHAIN_API_KEY=<your-api-key>
+export LANGCHAIN_PROJECT=<your-project>  # if not specified, defaults to "default"
+```
+
+If you are inside this directory, then you can spin up a LangServe instance directly by:
+
+```shell
+langchain serve
+```
+
+This will start the FastAPI app with a server is running locally at
+[http://localhost:8000](http://localhost:8000)
+
+We can see all templates at [http://127.0.0.1:8000/docs](http://127.0.0.1:8000/docs)
+We can access the playground at [http://127.0.0.1:8000/rag-chroma-dense-retrieval/playground](http://127.0.0.1:8000/rag-chroma-dense-retrieval/playground)
+
+We can access the template from code with:
+
+```python
+from langserve.client import RemoteRunnable
+
+runnable = RemoteRunnable("http://localhost:8000/rag-chroma-dense-retrieval")
+```

templates/rag-chroma-dense-retrieval/pyproject.toml

@@ -0,0 +1,35 @@
+[tool.poetry]
+name = "rag-chroma-dense-retrieval"
+version = "0.1.0"
+description = "Dense retrieval using vectorized propositions.s"
+authors = [
+    "William Fu-Hinthorn <[email protected]>",
+]
+readme = "README.md"
+
+[tool.poetry.dependencies]
+python = ">=3.8.1,<4.0"
+langchain = ">=0.0.350"
+openai = "<2"
+tiktoken = ">=0.5.1"
+chromadb = ">=0.4.14"
+bs4 = "^0.0.1"
+
+[tool.poetry.group.dev.dependencies]
+langchain-cli = ">=0.0.15"
+
+[tool.langserve]
+export_module = "rag_chroma_multi_modal_multi_vector"
+export_attr = "chain"
+
+[tool.templates-hub]
+use-case = "rag"
+author = "LangChain"
+integrations = ["OpenAI", "Chroma"]
+tags = ["vectordbs"]
+
+[build-system]
+requires = [
+    "poetry-core",
+]
+build-backend = "poetry.core.masonry.api"

templates/rag-chroma-dense-retrieval/rag_chroma_dense_retrieval.ipynb

@@ -0,0 +1,68 @@
+{
+ "cells": [
+  {
+   "attachments": {},
+   "cell_type": "markdown",
+   "id": "681a5d1e",
+   "metadata": {},
+   "source": [
+    "## Run Template\n",
+    "\n",
+    "In `server.py`, set -\n",
+    "```\n",
+    "from fastapi import FastAPI\n",
+    "from langserve import add_routes\n",
+    "from rag_chroma_dense_retrieval import chain\n",
+    "\n",
+    "app = FastAPI(\n",
+    "    title=\"LangChain Server\",\n",
+    "    version=\"1.0\",\n",
+    "    description=\"Retriever and Generator for RAG Chroma Dense Retrieval\",\n",
+    ")\n",
+    "\n",
+    "add_routes(app, chain, path=\"/rag-chroma-dense-retrieval\")\n",
+    "\n",
+    "if __name__ == \"__main__\":\n",
+    "    import uvicorn\n",
+    "\n",
+    "    uvicorn.run(app, host=\"localhost\", port=8000)\n",
+    "\n",
+    "```"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "d774be2a",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from langserve.client import RemoteRunnable\n",
+    "\n",
+    "rag_app = RemoteRunnable(\"http://localhost:8001/rag-chroma-dense-retrieval\")\n",
+    "rag_app.invoke(\"How are transformers related to convolutional neural networks?\")"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.11.2"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

templates/rag-chroma-dense-retrieval/rag_chroma_dense_retrieval/__init__.py

@@ -0,0 +1,4 @@
+from rag_chroma_dense_retrieval.chain import chain
+from rag_chroma_dense_retrieval.proposal_chain import proposition_chain
+
+__all__ = ["chain", "proposition_chain"]

templates/rag-chroma-dense-retrieval/rag_chroma_dense_retrieval/chain.py

@@ -0,0 +1,67 @@
+from langchain_community.chat_models import ChatOpenAI
+from langchain_core.load import load
+from langchain_core.output_parsers import StrOutputParser
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_core.pydantic_v1 import BaseModel
+from langchain_core.runnables import RunnablePassthrough
+
+from rag_chroma_dense_retrieval.constants import DOCSTORE_ID_KEY
+from rag_chroma_dense_retrieval.storage import get_multi_vector_retriever
+
+
+def format_docs(docs: list) -> str:
+    loaded_docs = [load(doc) for doc in docs]
+    return "\n".join(
+        [
+            f"<Document id={i}>\n{doc.page_content}\n</Document>"
+            for i, doc in enumerate(loaded_docs)
+        ]
+    )
+
+
+def rag_chain(retriever):
+    """
+    The RAG chain
+
+    :param retriever: A function that retrieves the necessary context for the model.
+    :return: A chain of functions representing the multi-modal RAG process.
+    """
+    model = ChatOpenAI(temperature=0, model="gpt-4-1106-preview", max_tokens=1024)
+    prompt = ChatPromptTemplate.from_messages(
+        [
+            (
+                "system",
+                "You are an AI assistant. Answer based on the retrieved documents:"
+                "\n<Documents>\n{context}\n</Documents>",
+            ),
+            ("user", "{question}?"),
+        ]
+    )
+
+    # Define the RAG pipeline
+    chain = (
+        {
+            "context": retriever | format_docs,
+            "question": RunnablePassthrough(),
+        }
+        | prompt
+        | model
+        | StrOutputParser()
+    )
+
+    return chain
+
+
+# Create the multi-vector retriever
+retriever = get_multi_vector_retriever(DOCSTORE_ID_KEY)
+
+# Create RAG chain
+chain = rag_chain(retriever)
+
+
+# Add typing for input
+class Question(BaseModel):
+    __root__: str
+
+
+chain = chain.with_types(input_type=Question)

templates/rag-chroma-dense-retrieval/rag_chroma_dense_retrieval/constants.py

@@ -0,0 +1 @@
+DOCSTORE_ID_KEY = "doc_id"

templates/rag-chroma-dense-retrieval/rag_chroma_dense_retrieval/ingest.py

@@ -0,0 +1,87 @@
+import logging
+import uuid
+from typing import Sequence
+
+from langchain_core.documents import Document
+from langchain_core.runnables import Runnable
+
+from rag_chroma_dense_retrieval.constants import DOCSTORE_ID_KEY
+from rag_chroma_dense_retrieval.proposal_chain import proposition_chain
+from rag_chroma_dense_retrieval.storage import get_multi_vector_retriever
+
+logging.basicConfig(level=logging.INFO)
+
+logger = logging.getLogger(__name__)
+
+
+def add_documents(
+    retriever,
+    propositions: Sequence[Sequence[str]],
+    docs: Sequence[Document],
+    id_key: str = DOCSTORE_ID_KEY,
+):
+    doc_ids = [
+        str(uuid.uuid5(uuid.NAMESPACE_DNS, doc.metadata["source"])) for doc in docs
+    ]
+    prop_docs = [
+        Document(page_content=prop, metadata={id_key: doc_ids[i]})
+        for i, props in enumerate(propositions)
+        for prop in props
+        if prop
+    ]
+    retriever.vectorstore.add_documents(prop_docs)
+    retriever.docstore.mset(list(zip(doc_ids, docs)))
+
+
+def create_index(
+    docs: Sequence[Document],
+    indexer: Runnable,
+    docstore_id_key: str = DOCSTORE_ID_KEY,
+):
+    """
+    Create retriever that indexes docs and their propositions
+
+    :param docs: Documents to index
+    :param indexer: Runnable creates additional propositions per doc
+    :param docstore_id_key: Key to use to store the docstore id
+    :return: Retriever
+    """
+    logger.info("Creating multi-vector retriever")
+    retriever = get_multi_vector_retriever(docstore_id_key)
+    propositions = indexer.batch([{"input": doc.page_content} for doc in docs])
+
+    add_documents(
+        retriever,
+        propositions,
+        docs,
+        id_key=docstore_id_key,
+    )
+
+    return retriever
+
+
+if __name__ == "__main__":
+    # For our example, we'll load docs from the web
+    from langchain.text_splitter import RecursiveCharacterTextSplitter  # noqa
+    from langchain_community.document_loaders.recursive_url_loader import (
+        RecursiveUrlLoader,
+    )  # noqa
+
+    # The attention is all you need paper
+    # Could add more parsing here, as it's very raw.
+    loader = RecursiveUrlLoader("https://ar5iv.labs.arxiv.org/html/1706.03762")
+    data = loader.load()
+    logger.info(f"Loaded {len(data)} documents")
+
+    # Split
+
+    text_splitter = RecursiveCharacterTextSplitter(chunk_size=8000, chunk_overlap=0)
+    all_splits = text_splitter.split_documents(data)
+    logger.info(f"Split into {len(all_splits)} documents")
+
+    # Create retriever
+    retriever_multi_vector_img = create_index(
+        all_splits,
+        proposition_chain,
+        DOCSTORE_ID_KEY,
+    )