templates: Add NVIDIA Canonical RAG example chain (#15758)

- **Description:** Adds a RAG template that uses NVIDIA AI playground
and embedding models, along with Milvus vector store

- **Dependencies:** This template depends on the AI playground service
in NVIDIA NGC. API keys with a significant trial compute are available
(10k queries at the time of writing). This template also depends on the
Milvus Vector store which is publicly available.

Note: [A quick link to get a
key](https://catalog.ngc.nvidia.com/orgs/nvidia/teams/ai-foundation/models/codellama-13b/api)
when you have an NGC account. Generate Key button at the top right of
the code window.

---------

Co-authored-by: Sagar B Manjunath <sbogadimanju@nvidia.com>
Co-authored-by: Erick Friis <erick@langchain.dev>

templates/nvidia-rag-canonical/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/nvidia-rag-canonical/README.md

@@ -0,0 +1,121 @@
+
+# nvidia-rag-canonical
+
+This template performs RAG using Milvus Vector Store and NVIDIA Models (Embedding and Chat).
+
+## Environment Setup
+
+You should export your NVIDIA API Key as an environment variable.
+If you do not have an NVIDIA API Key, you can create one by following these steps:
+1. Create a free account with the [NVIDIA GPU Cloud](https://catalog.ngc.nvidia.com/) service, which hosts AI solution catalogs, containers, models, etc.
+2. Navigate to `Catalog > AI Foundation Models > (Model with API endpoint)`.
+3. Select the `API` option and click `Generate Key`.
+4. Save the generated key as `NVIDIA_API_KEY`. From there, you should have access to the endpoints.
+
+```shell
+export NVIDIA_API_KEY=...
+```
+
+For instructions on hosting the Milvus Vector Store, refer to the section at the bottom.
+
+## Usage
+
+To use this package, you should first have the LangChain CLI installed:
+
+```shell
+pip install -U langchain-cli
+```
+
+To use the NVIDIA models, install the Langchain NVIDIA AI Endpoints package:
+```shell
+pip install -U langchain_nvidia_aiplay
+```
+
+To create a new LangChain project and install this as the only package, you can do:
+
+```shell
+langchain app new my-app --package nvidia-rag-canonical
+```
+
+If you want to add this to an existing project, you can just run:
+
+```shell
+langchain app add nvidia-rag-canonical
+```
+
+And add the following code to your `server.py` file:
+```python
+from nvidia_rag_canonical import chain as rag_nvidia_chain
+
+add_routes(app, rag_nvidia_chain, path="/nvidia-rag")
+```
+
+If you want to set up an ingestion pipeline, you can add the following code to your `server.py` file:
+```python
+from rag_nvidia_canonical import ingest as rag_nvidia_ingest
+
+add_routes(app, rag_nvidia_ingest, path="/nvidia-rag-ingest")
+```
+Note that for files ingested by the ingestion API, the server will need to be restarted for the newly ingested files to be accessible by the retriever.
+
+(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 DO NOT already have a Milvus Vector Store you want to connect to, see `Milvus Setup` section below before proceeding.
+
+If you DO have a Milvus Vector Store you want to connect to, edit the connection details in `nvidia_rag_canonical/chain.py`
+
+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/nvidia-rag/playground](http://127.0.0.1:8000/nvidia-rag/playground)
+
+We can access the template from code with:
+
+```python
+from langserve.client import RemoteRunnable
+
+runnable = RemoteRunnable("http://localhost:8000/nvidia-rag")
+```
+
+
+## Milvus Setup
+
+Use this step if you need to create a Milvus Vector Store and ingest data.
+We will first follow the standard Milvus setup instructions [here](https://milvus.io/docs/install_standalone-docker.md).
+
+1. Download the Docker Compose YAML file.
+    ```shell
+    wget https://github.com/milvus-io/milvus/releases/download/v2.3.3/milvus-standalone-docker-compose.yml -O docker-compose.yml
+    ```
+2. Start the Milvus Vector Store container
+    ```shell
+    sudo docker compose up -d
+    ```
+3. Install the PyMilvus package to interact with the Milvus container.
+    ```shell
+    pip install pymilvus
+    ```
+4. Let's now ingest some data! We can do that by moving into this directory and running the code in `ingest.py`, eg:
+
+    ```shell
+    python ingest.py
+    ```
+
+    Note that you can (and should!) change this to ingest data of your choice.

templates/nvidia-rag-canonical/ingest.py

@@ -0,0 +1,39 @@
+import getpass
+import os
+
+from langchain.document_loaders import PyPDFLoader
+from langchain.text_splitter import CharacterTextSplitter
+from langchain.vectorstores.milvus import Milvus
+from langchain_nvidia_aiplay import NVIDIAEmbeddings
+
+if os.environ.get("NVIDIA_API_KEY", "").startswith("nvapi-"):
+    print("Valid NVIDIA_API_KEY already in environment. Delete to reset")
+else:
+    nvapi_key = getpass.getpass("NVAPI Key (starts with nvapi-): ")
+    assert nvapi_key.startswith("nvapi-"), f"{nvapi_key[:5]}... is not a valid key"
+    os.environ["NVIDIA_API_KEY"] = nvapi_key
+
+# Note: if you change this, you should also change it in `nvidia_rag_canonical/chain.py`
+EMBEDDING_MODEL = "nvolveqa_40k"
+HOST = "127.0.0.1"
+PORT = "19530"
+COLLECTION_NAME = "test"
+
+embeddings = NVIDIAEmbeddings(model=EMBEDDING_MODEL)
+
+if __name__ == "__main__":
+    # Load docs
+    loader = PyPDFLoader("https://www.ssa.gov/news/press/factsheets/basicfact-alt.pdf")
+    data = loader.load()
+
+    # Split docs
+    text_splitter = CharacterTextSplitter(chunk_size=300, chunk_overlap=100)
+    docs = text_splitter.split_documents(data)
+
+    # Insert the documents in Milvus Vector Store
+    vector_db = Milvus.from_documents(
+        docs,
+        embeddings,
+        collection_name=COLLECTION_NAME,
+        connection_args={"host": HOST, "port": PORT},
+    )

templates/nvidia-rag-canonical/nvidia_rag_canonical/__init__.py

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

templates/nvidia-rag-canonical/nvidia_rag_canonical/chain.py

@@ -0,0 +1,91 @@
+import getpass
+import os
+
+from langchain.text_splitter import CharacterTextSplitter
+from langchain_community.document_loaders import PyPDFLoader
+from langchain_community.vectorstores import Milvus
+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 (
+    RunnableLambda,
+    RunnableParallel,
+    RunnablePassthrough,
+)
+from langchain_nvidia_aiplay import ChatNVIDIA, NVIDIAEmbeddings
+
+EMBEDDING_MODEL = "nvolveqa_40k"
+CHAT_MODEL = "llama2_13b"
+HOST = "127.0.0.1"
+PORT = "19530"
+COLLECTION_NAME = "test"
+INGESTION_CHUNK_SIZE = 500
+INGESTION_CHUNK_OVERLAP = 0
+
+if os.environ.get("NVIDIA_API_KEY", "").startswith("nvapi-"):
+    print("Valid NVIDIA_API_KEY already in environment. Delete to reset")
+else:
+    nvapi_key = getpass.getpass("NVAPI Key (starts with nvapi-): ")
+    assert nvapi_key.startswith("nvapi-"), f"{nvapi_key[:5]}... is not a valid key"
+    os.environ["NVIDIA_API_KEY"] = nvapi_key
+
+# Read from Milvus Vector Store
+embeddings = NVIDIAEmbeddings(model=EMBEDDING_MODEL)
+vectorstore = Milvus(
+    connection_args={"host": HOST, "port": PORT},
+    collection_name=COLLECTION_NAME,
+    embedding_function=embeddings,
+)
+retriever = vectorstore.as_retriever()
+
+# RAG prompt
+template = """<s>[INST] <<SYS>>
+Use the following context to answer the user's question. If you don't know the answer,
+just say that you don't know, don't try to make up an answer.
+<</SYS>>
+<s>[INST] Context: {context} Question: {question} Only return the helpful
+ answer below and nothing else. Helpful answer:[/INST]"
+"""
+prompt = ChatPromptTemplate.from_template(template)
+
+# RAG
+model = ChatNVIDIA(model=CHAT_MODEL)
+chain = (
+    RunnableParallel({"context": retriever, "question": RunnablePassthrough()})
+    | prompt
+    | model
+    | StrOutputParser()
+)
+
+
+# Add typing for input
+class Question(BaseModel):
+    __root__: str
+
+
+chain = chain.with_types(input_type=Question)
+
+
+def _ingest(url: str) -> dict:
+    """Load and ingest the PDF file from the URL"""
+
+    loader = PyPDFLoader(url)
+    data = loader.load()
+
+    # Split docs
+    text_splitter = CharacterTextSplitter(
+        chunk_size=INGESTION_CHUNK_SIZE, chunk_overlap=INGESTION_CHUNK_OVERLAP
+    )
+    docs = text_splitter.split_documents(data)
+
+    # Insert the documents in Milvus Vector Store
+    _ = Milvus.from_documents(
+        documents=docs,
+        embedding=embeddings,
+        collection_name=COLLECTION_NAME,
+        connection_args={"host": HOST, "port": PORT},
+    )
+    return {}
+
+
+ingest = RunnableLambda(_ingest)

templates/nvidia-rag-canonical/pyproject.toml

@@ -0,0 +1,29 @@
+[tool.poetry]
+name = "nvidia-rag-canonical"
+version = "0.1.0"
+description = "RAG with NVIDIA"
+authors = ["Sagar Bogadi Manjunath <sbogadimanju@nvidia.com>"]
+readme = "README.md"
+
+[tool.poetry.dependencies]
+python = ">=3.8.1,<4.0"
+langchain = "^0.1"
+pymilvus = ">=2.3.0"
+langchain-nvidia-aiplay = "^0.0.2"
+
+[tool.poetry.group.dev.dependencies]
+langchain-cli = ">=0.0.20"
+
+[tool.langserve]
+export_module = "nvidia_rag_canonical"
+export_attr = "chain"
+
+[tool.templates-hub]
+use-case = "rag"
+author = "LangChain"
+integrations = ["Milvus", "NVIDIA"]
+tags = ["vectordbs"]
+
+[build-system]
+requires = ["poetry-core"]
+build-backend = "poetry.core.masonry.api"