templates[patch]: Add cohere librarian template (#14601)

Adding the example I build for the Cohere hackathon.

It can:

use a vector database to reccommend books

<img width="840" alt="image"
src="https://github.com/langchain-ai/langchain/assets/144115527/96543a18-217b-4445-ab4b-950c7cced915">

Use a prompt template to provide information about the library

<img width="834" alt="image"
src="https://github.com/langchain-ai/langchain/assets/144115527/996c8e0f-cab0-4213-bcc9-9baf84f1494b">

Use Cohere RAG to provide grounded results

<img width="822" alt="image"
src="https://github.com/langchain-ai/langchain/assets/144115527/7bb4a883-5316-41a9-9d2e-19fd49a43dcb">

---------

Co-authored-by: Erick Friis <[email protected]>

templates/cohere-librarian/README.md

@@ -0,0 +1,71 @@
+
+# cohere-librarian
+
+This template turns Cohere into a librarian.
+
+It demonstrates the use of a router to switch between chains that can handle different things: a vector database with Cohere embeddings; a chat bot that has a prompt with some information about the library; and finally a RAG chatbot that has access to the internet.
+
+For a fuller demo of the book recomendation, consider replacing books_with_blurbs.csv with a larger sample from the following dataset: https://www.kaggle.com/datasets/jdobrow/57000-books-with-metadata-and-blurbs/ .
+
+## Environment Setup
+
+Set the `COHERE_API_KEY` environment variable to access the Cohere models.
+
+## 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 cohere-librarian
+```
+
+If you want to add this to an existing project, you can just run:
+
+```shell
+langchain app add cohere-librarian
+```
+
+And add the following code to your `server.py` file:
+```python
+from cohere_librarian.chain import chain as cohere_librarian_chain
+
+add_routes(app, cohere_librarian_chain, path="/cohere-librarian")
+```
+
+(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://localhost:8000/docs](http://localhost:8000/docs)
+We can access the playground at [http://localhost:8000/cohere-librarian/playground](http://localhost:8000/cohere-librarian/playground)  
+
+We can access the template from code with:
+
+```python
+from langserve.client import RemoteRunnable
+
+runnable = RemoteRunnable("http://localhost:8000/cohere-librarian")
+```

templates/cohere-librarian/cohere_librarian/__init__.py

@@ -0,0 +1,3 @@
+from .chain import chain
+
+__all__ = ["chain"]

templates/cohere-librarian/cohere_librarian/blurb_matcher.py

@@ -0,0 +1,49 @@
+import csv
+
+from langchain.chains.question_answering import load_qa_chain
+from langchain.embeddings import CohereEmbeddings
+from langchain.prompts import PromptTemplate
+from langchain.vectorstores import Chroma
+
+from .chat import chat
+
+csv_file = open("data/books_with_blurbs.csv", "r")
+csv_reader = csv.reader(csv_file)
+csv_data = list(csv_reader)
+parsed_data = [
+    {
+        "id": x[0],
+        "title": x[1],
+        "author": x[2],
+        "year": x[3],
+        "publisher": x[4],
+        "blurb": x[5],
+    }
+    for x in csv_data
+]
+parsed_data[1]
+
+embeddings = CohereEmbeddings()
+
+docsearch = Chroma.from_texts(
+    [x["title"] for x in parsed_data], embeddings, metadatas=parsed_data
+).as_retriever()
+
+
+prompt_template = """
+{context}
+
+Use the book reccommendations to suggest books for the user to read.
+Only use the titles of the books, do not make up titles. Format the response as
+a bulleted list prefixed by a relevant message.
+
+User: {message}"""
+
+PROMPT = PromptTemplate(
+    template=prompt_template, input_variables=["context", "message"]
+)
+
+book_rec_chain = {
+    "input_documents": lambda x: docsearch.get_relevant_documents(x["message"]),
+    "message": lambda x: x["message"],
+} | load_qa_chain(chat, chain_type="stuff", prompt=PROMPT)

templates/cohere-librarian/cohere_librarian/chain.py

@@ -0,0 +1,10 @@
+from langchain.pydantic_v1 import BaseModel
+
+from .router import branched_chain
+
+
+class ChainInput(BaseModel):
+    message: str
+
+
+chain = branched_chain.with_types(input_type=ChainInput)

templates/cohere-librarian/cohere_librarian/chat.py

@@ -0,0 +1,3 @@
+from langchain.llms import Cohere
+
+chat = Cohere()

templates/cohere-librarian/cohere_librarian/library_info.py

@@ -0,0 +1,27 @@
+from langchain.prompts import (
+    ChatPromptTemplate,
+    HumanMessagePromptTemplate,
+    SystemMessagePromptTemplate,
+)
+
+from .chat import chat
+
+librarian_prompt = ChatPromptTemplate.from_messages(
+    [
+        SystemMessagePromptTemplate.from_template(
+            """
+            You are a librarian at cohere community library. Your job is to
+            help recommend people books to read based on their interests and
+            preferences. You also give information about the library.
+
+            The library opens at 8am and closes at 9pm daily. It is closed on
+            Sundays.
+
+            Please answer the following message:
+            """
+        ),
+        HumanMessagePromptTemplate.from_template("{message}"),
+    ]
+)
+
+library_info = librarian_prompt | chat

templates/cohere-librarian/cohere_librarian/rag.py

@@ -0,0 +1,16 @@
+from langchain.chat_models import ChatCohere
+from langchain.retrievers import CohereRagRetriever
+
+rag = CohereRagRetriever(llm=ChatCohere())
+
+
+def get_docs_message(message):
+    docs = rag.get_relevant_documents(message)
+    message_doc = next(
+        (x for x in docs if x.metadata.get("type") == "model_response"), None
+    )
+    return message_doc.page_content
+
+
+def librarian_rag(x):
+    return get_docs_message(x["message"])

templates/cohere-librarian/cohere_librarian/router.py

@@ -0,0 +1,43 @@
+from langchain.prompts import ChatPromptTemplate
+from langchain.schema.output_parser import StrOutputParser
+from langchain.schema.runnable import RunnableBranch
+
+from .blurb_matcher import book_rec_chain
+from .chat import chat
+from .library_info import library_info
+from .rag import librarian_rag
+
+chain = (
+    ChatPromptTemplate.from_template(
+        """Given the user message below,
+classify it as either being about `recommendation`, `library` or `other`.
+
+'{message}'
+
+Respond with just one word.
+For example, if the message is about a book recommendation,respond with 
+`recommendation`.
+"""
+    )
+    | chat
+    | StrOutputParser()
+)
+
+
+def extract_op_field(x):
+    return x["output_text"]
+
+
+branch = RunnableBranch(
+    (
+        lambda x: "recommendation" in x["topic"].lower(),
+        book_rec_chain | extract_op_field,
+    ),
+    (
+        lambda x: "library" in x["topic"].lower(),
+        {"message": lambda x: x["message"]} | library_info,
+    ),
+    librarian_rag,
+)
+
+branched_chain = {"topic": chain, "message": lambda x: x["message"]} | branch