RAG Optimization

Dockerfile

@@ -0,0 +1,12 @@
+FROM python:3.9
+RUN mkdir -p $HOME/app/data/vectorstore && chown -R user:user $HOME/app/data and in app.py via constant path to make a directory path: DATA_DIR = "./data" VECTORSTORE_DIR = os.path.join(DATA_DIR, "vectorstore") VECTORSTORE_PATH = os.path.join(VECTORSTORE_DIR, "index.faiss")
+RUN useradd -m -u 1000 user
+USER user
+ENV HOME=/home/user \
+    PATH=/home/user/.local/bin:$PATH
+WORKDIR $HOME/app
+COPY --chown=user . $HOME/app
+COPY ./requirements.txt ~/app/requirements.txt
+RUN pip install -r requirements.txt
+COPY . .
+CMD ["chainlit", "run", "app.py", "--port", "7860"]

README.md

@@ -1,10 +1,190 @@
 ---
-title: RAG Optimization
-emoji: 📈
-colorFrom: red
-colorTo: red
+title: BeyondChatGPT Demo
+emoji: 📉
+colorFrom: pink
+colorTo: yellow
 sdk: docker
 pinned: false
+app_port: 7860
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+<p align = "center" draggable=”false” ><img src="https://github.com/AI-Maker-Space/LLM-Dev-101/assets/37101144/d1343317-fa2f-41e1-8af1-1dbb18399719" 
+     width="200px"
+     height="auto"/>
+</p>
+
+
+## <h1 align="center" id="heading">:wave: Welcome to Beyond ChatGPT!!</h1>
+
+For a step-by-step YouTube video walkthrough, watch this! [Deploying Chainlit app on Hugging Face](https://www.youtube.com/live/pRbbZcL0NMI?si=NAYhMZ_suAY84f06&t=2119)
+
+![Beyond ChatGPT: Build Your First LLM Application](https://github.com/AI-Maker-Space/Beyond-ChatGPT/assets/48775140/cb7a74b8-28af-4d12-a008-8f5a51d47b4c)
+
+## 🤖 Your First LLM App
+
+> If you need an introduction to `git`, or information on how to set up API keys for the tools we'll be using in this repository - check out our [Interactive Dev Environment for LLM Development](https://github.com/AI-Maker-Space/Interactive-Dev-Environment-for-LLM-Development/tree/main) which has everything you'd need to get started in this repository!
+
+In this repository, we'll walk you through the steps to create a Large Language Model (LLM) application using Chainlit, then containerize it using Docker, and finally deploy it on Huggingface Spaces.
+
+Are you ready? Let's get started!
+
+<details>
+  <summary>🖥️ Accessing "gpt-3.5-turbo" (ChatGPT) like a developer</summary>
+
+1. Head to [this notebook](https://colab.research.google.com/drive/1mOzbgf4a2SP5qQj33ZxTz2a01-5eXqk2?usp=sharing) and follow along with the instructions!
+
+2. Complete the notebook and try out your own system/assistant messages!
+
+That's it! Head to the next step and start building your application!
+
+</details>
+
+
+<details>
+  <summary>🏗️ Building Your First LLM App</summary>
+
+1. Clone [this](https://github.com/AI-Maker-Space/Beyond-ChatGPT/tree/main) repo.
+
+     ``` bash
+     git clone https://github.com/AI-Maker-Space/Beyond-ChatGPT.git
+     ```
+
+2. Navigate inside this repo
+     ``` bash
+     cd Beyond-ChatGPT
+     ```
+
+3. Install the packages required for this python envirnoment in `requirements.txt`.
+     ``` bash
+     pip install -r requirements.txt
+     ``` 
+
+4. Open your `.env` file. Replace the `###` in your `.env` file with your OpenAI Key and save the file.
+     ``` bash
+     OPENAI_API_KEY=sk-###
+     ```
+
+5. Let's try deploying it locally. Make sure you're in the python environment where you installed Chainlit and OpenAI. Run the app using Chainlit. This may take a minute to run.
+     ``` bash
+     chainlit run app.py -w
+     ```
+
+<p align = "center" draggable=”false”>
+<img src="https://github.com/AI-Maker-Space/LLMOps-Dev-101/assets/37101144/54bcccf9-12e2-4cef-ab53-585c1e2b0fb5"> 
+</p>
+
+Great work! Let's see if we can interact with our chatbot.
+
+<p align = "center" draggable=”false”>
+<img src="https://github.com/AI-Maker-Space/LLMOps-Dev-101/assets/37101144/854e4435-1dee-438a-9146-7174b39f7c61"> 
+</p> 
+
+Awesome! Time to throw it into a docker container and prepare it for shipping!
+</details>
+
+
+
+<details>
+  <summary>🐳 Containerizing our App</summary>
+
+1. Let's build the Docker image. We'll tag our image as `llm-app` using the `-t` parameter. The `.` at the end means we want all of the files in our current directory to be added to our image.
+     
+     ``` bash
+     docker build -t llm-app .
+     ```
+
+2. Run and test the Docker image locally using the `run` command. The `-p`parameter connects our **host port #** to the left of the `:` to our **container port #** on the right.
+    
+     ``` bash
+     docker run -p 7860:7860 llm-app
+     ```
+
+3. Visit http://localhost:7860 in your browser to see if the app runs correctly.
+
+<p align = "center" draggable=”false”>
+<img src="https://github.com/AI-Maker-Space/LLMOps-Dev-101/assets/37101144/2c764f25-09a0-431b-8d28-32246e0ca1b7"> 
+</p>
+
+Great! Time to ship!
+</details>
+
+
+<details>
+  <summary>🚀 Deploying Your First LLM App</summary>
+
+1. Let's create a new Huggingface Space. Navigate to [Huggingface](https://huggingface.co) and click on your profile picture on the top right. Then click on `New Space`.
+
+<p align = "center" draggable=”false”>
+<img src="https://github.com/AI-Maker-Space/LLMOps-Dev-101/assets/37101144/f0656408-28b8-4876-9887-8f0c4b882bae"> 
+</p>
+
+2. Setup your space as shown below:
+   
+- Owner: Your username
+- Space Name: `llm-app`
+- License: `Openrail`
+- Select the Space SDK: `Docker`
+- Docker Template: `Blank`
+- Space Hardware: `CPU basic - 2 vCPU - 16 GB - Free`
+- Repo type: `Public`
+
+<p align = "center" draggable=”false”>
+<img src="https://github.com/AI-Maker-Space/LLMOps-Dev-101/assets/37101144/8f16afd1-6b46-4d9f-b642-8fefe355c5c9"> 
+</p>
+
+3. You should see something like this. We're now ready to send our files to our Huggingface Space. After cloning, move your files to this repo and push it along with your docker file. You DO NOT need to create a Dockerfile. Make sure NOT TO push your `.env` file. This should automatically be ignored.
+
+<p align = "center" draggable=”false”>
+<img src="https://github.com/AI-Maker-Space/LLMOps-Dev-101/assets/37101144/cbf366e2-7613-4223-932a-72c67a73f9c6"> 
+</p>
+
+4. After pushing all files, navigate to the settings in the top right to add your OpenAI API key.
+
+<p align = "center" draggable=”false”>
+<img src="https://github.com/AI-Maker-Space/LLMOps-Dev-101/assets/37101144/a1123a6f-abdd-4f76-bea4-39acf9928762"> 
+</p>
+
+5. Scroll down to `Variables and secrets` and click on `New secret` on the top right.
+
+<p align = "center" draggable=”false”>
+<img src="https://github.com/AI-Maker-Space/LLMOps-Dev-101/assets/37101144/a8a4a25d-752b-4036-b572-93381370c2db"> 
+</p>
+
+6. Set the name to `OPENAI_API_KEY` and add your OpenAI key under `Value`. Click save.
+
+<p align = "center" draggable=”false”>
+<img src="https://github.com/AI-Maker-Space/LLMOps-Dev-101/assets/37101144/0a897538-1779-48ff-bcb4-486af30f7a14"> 
+</p>
+
+7. To ensure your key is being used, we recommend you `Restart this Space`.
+
+<p align = "center" draggable=”false”>
+<img src="https://github.com/AI-Maker-Space/LLMOps-Dev-101/assets/37101144/fb1d83af-6ebe-4676-8bf5-b6d88f07c583"> 
+</p>
+
+8. Congratulations! You just deployed your first LLM! 🚀🚀🚀 Get on linkedin and post your results and experience! Make sure to tag us at #AIMakerspace !
+
+Here's a template to get your post started!
+
+```
+🚀🎉 Exciting News! 🎉🚀
+
+🏗️ Today, I'm thrilled to announce that I've successfully built and shipped my first-ever LLM using the powerful combination of Chainlit, Docker, and the OpenAI API! 🖥️
+
+Check it out 👇
+[LINK TO APP]
+
+A big shoutout to the @**AI Makerspace** for all making this possible. Couldn't have done it without the incredible community there. 🤗🙏
+
+Looking forward to building with the community! 🙌✨ Here's to many more creations ahead! 🥂🎉
+
+Who else is diving into the world of AI? Let's connect! 🌐💡
+
+#FirstLLM #Chainlit #Docker #OpenAI #AIMakerspace
+```
+
+</details>
+
+<p></p>
+
+### That's it for now!  And so it begins.... :)

app.py

@@ -0,0 +1,152 @@
+import os
+import chainlit as cl
+from dotenv import load_dotenv
+from operator import itemgetter
+from langchain_huggingface import HuggingFaceEndpoint
+from langchain_community.document_loaders import TextLoader
+from langchain_text_splitters import RecursiveCharacterTextSplitter
+from langchain_community.vectorstores import FAISS
+from langchain_huggingface import HuggingFaceEndpointEmbeddings
+from langchain_core.prompts import PromptTemplate
+from langchain.schema.output_parser import StrOutputParser
+from langchain.schema.runnable import RunnablePassthrough
+from langchain.schema.runnable.config import RunnableConfig
+
+# GLOBAL SCOPE - ENTIRE APPLICATION HAS ACCESS TO VALUES SET IN THIS SCOPE #
+# ---- ENV VARIABLES ---- # 
+"""
+This function will load our environment file (.env) if it is present.
+
+NOTE: Make sure that .env is in your .gitignore file - it is by default, but please ensure it remains there.
+"""
+load_dotenv()
+
+"""
+We will load our environment variables here.
+"""
+HF_LLM_ENDPOINT = os.environ["HF_LLM_ENDPOINT"]
+HF_EMBED_ENDPOINT = os.environ["HF_EMBED_ENDPOINT"]
+HF_TOKEN = os.environ["HF_TOKEN"]
+
+# ---- GLOBAL DECLARATIONS ---- #
+
+# -- RETRIEVAL -- #
+"""
+1. Load Documents from Text File
+2. Split Documents into Chunks
+3. Load HuggingFace Embeddings (remember to use the URL we set above)
+4. Index Files if they do not exist, otherwise load the vectorstore
+"""
+document_loader = TextLoader("./data/paul_graham_essays.txt")
+documents = document_loader.load()
+
+text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=30)
+split_documents = text_splitter.split_documents(documents)
+
+hf_embeddings = HuggingFaceEndpointEmbeddings(
+    model=HF_EMBED_ENDPOINT,
+    task="feature-extraction",
+    huggingfacehub_api_token=HF_TOKEN,
+)
+
+if os.path.exists("./data/vectorstore"):
+    vectorstore = FAISS.load_local(
+        "./data/vectorstore", 
+        hf_embeddings, 
+        allow_dangerous_deserialization=True # this is necessary to load the vectorstore from disk as it's stored as a `.pkl` file.
+    )
+    hf_retriever = vectorstore.as_retriever()
+    print("Loaded Vectorstore")
+else:
+    print("Indexing Files")
+    os.makedirs("./data/vectorstore", exist_ok=True)
+    for i in range(0, len(split_documents), 32):
+        if i == 0:
+            vectorstore = FAISS.from_documents(split_documents[i:i+32], hf_embeddings)
+            continue
+        vectorstore.add_documents(split_documents[i:i+32])
+    vectorstore.save_local("./data/vectorstore")
+
+hf_retriever = vectorstore.as_retriever()
+
+# -- AUGMENTED -- #
+"""
+1. Define a String Template
+2. Create a Prompt Template from the String Template
+"""
+RAG_PROMPT_TEMPLATE = """\
+<|start_header_id|>system<|end_header_id|>
+You are a helpful assistant. You answer user questions based on provided context. If you can't answer the question with the provided context, say you don't know.<|eot_id|>
+
+<|start_header_id|>user<|end_header_id|>
+User Query:
+{query}
+
+Context:
+{context}<|eot_id|>
+
+<|start_header_id|>assistant<|end_header_id|>
+"""
+
+rag_prompt = PromptTemplate.from_template(RAG_PROMPT_TEMPLATE)
+
+# -- GENERATION -- #
+"""
+1. Create a HuggingFaceEndpoint for the LLM
+"""
+hf_llm = HuggingFaceEndpoint(
+    endpoint_url=HF_LLM_ENDPOINT,
+    max_new_tokens=512,
+    top_k=10,
+    top_p=0.95,
+    temperature=0.3,
+    repetition_penalty=1.15,
+    huggingfacehub_api_token=HF_TOKEN,
+)
+
+@cl.author_rename
+def rename(original_author: str):
+    """
+    This function can be used to rename the 'author' of a message. 
+
+    In this case, we're overriding the 'Assistant' author to be 'Paul Graham Essay Bot'.
+    """
+    rename_dict = {
+        "Assistant" : "Paul Graham Essay Bot"
+    }
+    return rename_dict.get(original_author, original_author)
+
+@cl.on_chat_start
+async def start_chat():
+    """
+    This function will be called at the start of every user session. 
+
+    We will build our LCEL RAG chain here, and store it in the user session. 
+
+    The user session is a dictionary that is unique to each user session, and is stored in the memory of the server.
+    """
+
+    lcel_rag_chain = rag_prompt | hf_llm
+
+    cl.user_session.set("lcel_rag_chain", lcel_rag_chain)
+
+@cl.on_message  
+async def main(message: cl.Message):
+    """
+    This function will be called every time a message is recieved from a session.
+
+    We will use the LCEL RAG chain to generate a response to the user query.
+
+    The LCEL RAG chain is stored in the user session, and is unique to each user session - this is why we can access it here.
+    """
+    lcel_rag_chain = cl.user_session.get("lcel_rag_chain")
+
+    msg = cl.Message(content="")
+
+    async for chunk in lcel_rag_chain.astream(
+        {"query": message.content},
+        config=RunnableConfig(callbacks=[cl.LangchainCallbackHandler()]),
+    ):
+        await msg.stream_token(chunk)
+
+    await msg.send()

chainlit.md

@@ -0,0 +1 @@
+Ganesh's Optimized RAG

requirements.txt

@@ -0,0 +1,8 @@
+chainlit==0.7.700
+langchain==0.2.5
+langchain_community==0.2.5
+langchain_core==0.2.9
+langchain_huggingface==0.0.3
+langchain_text_splitters==0.2.1
+python-dotenv==1.0.1
+faiss-cpu