add the changes to app.py

.env.sample

@@ -0,0 +1,5 @@
+# !!! DO NOT UPDATE THIS FILE DIRECTLY. MAKE A COPY AND RENAME IT `.env` TO PROCEED !!! #
+HF_LLM_ENDPOINT="YOUR_LLM_ENDPOINT_URL_HERE"
+HF_EMBED_ENDPOINT="YOUR_EMBED_MODEL_ENDPOINT_URL_HERE"
+HF_TOKEN="YOUR_HF_TOKEN_HERE"
+# !!! DO NOT UPDATE THIS FILE DIRECTLY. MAKE A COPY AND RENAME IT `.env` TO PROCEED !!! #

.gitignore

@@ -0,0 +1,6 @@
+.env
+__pycache__/
+.chainlit
+*.faiss
+*.pkl
+.files

Dockerfile

@@ -0,0 +1,11 @@
+FROM python:3.9
+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 (2).md

@@ -0,0 +1,150 @@
+# Week 4: Tuesday
+
+In today's assignment, we'll be creating an Open Source LLM-powered LangChain RAG Application in Chainlit.
+
+There are 2 main sections to this assignment:
+
+## Build 🏗️
+
+### Build Task 1: Deploy LLM and Embedding Model to SageMaker Endpoint Through Hugging Face Inference Endpoints
+
+#### LLM Endpoint
+
+Select "Inference Endpoint" from the "Solutions" button in Hugging Face:
+
+![image](https://i.imgur.com/6KC9TCD.png)
+
+Create a "+ New Endpoint" from the Inference Endpoints dashboard.
+
+![image](https://i.imgur.com/G6Bq9KC.png)
+
+Select the `NousResearch/Meta-Llama-3-8B-Instruct` model repository and name your endpoint. Select N. Virginia as your region (`us-east-1`). Give your endpoint an appropriate name. Make sure to select *at least* a L4 GPU. 
+
+![image](https://i.imgur.com/X3YlUbh.png)
+
+Select the following settings for your `Advanced Configuration`.
+
+![image](https://i.imgur.com/c0HQ7g1.png)
+
+Create a `Protected` endpoint.
+
+![image](https://i.imgur.com/Ak8kchZ.png)
+
+If you were successful, you should see the following screen:
+
+![image](https://i.imgur.com/IBYG3wm.png)
+
+#### Embedding Model Endpoint
+We'll be using `Snowflake/snowflake-arctic-embed-m` for our embedding model today.
+
+The process is the same as the LLM - but we'll make a few specific tweaks:
+
+Let's make sure our set-up reflects the following screenshots:
+
+![image](https://i.imgur.com/IHh8FnC.png)
+
+After which, make sure the advanced configuration is set like so:
+
+![image](https://i.imgur.com/bbcrhUj.png)
+
+> #### NOTE: PLEASE SHUTDOWN YOUR INSTANCES WHEN YOU HAVE COMPLETED THE ASSIGNMENT TO PREVENT UNESSECARY CHARGES.
+
+### Build Task 2: Create RAG Pipeline with LangChain
+
+Follow the [notebook](https://colab.research.google.com/drive/1v1FYmvKH4gsqcdZwIT9wvbQe0GUjrc9d?usp=sharing) to create a LangChain pipeline powered by Hugging Face endpoints!
+
+Once you're done - please move on to Build Task 3!
+
+### Build Task 3: Create a Chainlit Application
+
+1. Create a new empty Docker space through Hugging Face - with the following settings:
+
+![image](https://i.imgur.com/0YzyQX7.png)
+
+> NOTE: You may notice the application builds slowly (~15min.) with the default free-tier hardware. The process will be faster using the `CPU upgrade` Space Hardware - though it is not required. 
+
+2. Clone the newly created space into a directory that is *NOT IN YOUR AI MAKERSPACE REPOSITORY* using the SSH option.
+
+> NOTE: You may need to ensure you've added your SSH key to Hugging Face, as well as GitHub. This should already be done.
+
+![image](https://i.imgur.com/5RyBdP5.png)
+
+3. Copy and Paste (`cp ...` or through UI) the contents of `Week 4/Day 1` into the newly cloned repository. 
+
+> NOTE: Please keep the `README.md` that was cloned from your space and delete the class `README.md`.
+
+4. Using the `ls` command or the `tree` command verify that you have copied over: 
+ - `app.py`
+ - `Dockerfile`
+ - `data/paul_graham_essays.txt`
+ - `chainlit.md`
+ - `.gitignore`
+ - `.env.sample`
+ - `solution_app.py`
+ - `requirements.txt`
+
+ Here is an example as the `ls -al` CLI command: 
+
+ ![image](https://i.imgur.com/vazGYeb.png)
+
+ 5. Work through the `app.py` file to migrate your LCEL LangChain RAG Chain from the Notebook to Chainlit!
+
+ 6. Be sure to modify your `README.md` and `chainlit.md` as you see fit!
+
+ > NOTE: If you get stuck, there is a working reference version in `solution_app.py`.
+
+ 7. When you are done with local testing - push your changes to your space. 
+
+ 8. Make sure you add your `HF_LLM_ENDPOINT`, `HF_EMBED_ENDPOINT`, `HF_TOKEN` as "Secrets" in your Hugging Face Space.
+
+### Terminating Your Resources
+
+Please head to the settings of each endpoint and select `Delete Endpoint`. You will need to type the name of the endpoint to delete the resources.
+
+### Deliverables
+
+- Completed Notebook
+- Chainlit Application in a Hugging Face Space Powered by Hugging Face Endpoints
+- Screenshot of endpoint usage
+
+Example Screen Shot:
+
+![image](https://i.imgur.com/qfbcVpS.png)
+
+## Ship 🚢
+
+Create a Hugging Face Space powered by Hugging Face Endpoints!
+
+### Deliverables
+
+- A short Loom of the space, and a 1min. walkthrough of the application in full
+
+## Share 🚀
+
+Make a social media post about your final application!
+
+### Deliverables
+
+- Make a post on any social media platform about what you built!
+
+Here's a template to get you started:
+
+```
+🚀 Exciting News! 🚀
+
+I am thrilled to announce that I have just built and shipped a open-source LLM-powered Retrieval Augmented Generation Application with LangChain! 🎉🤖
+
+🔍 Three Key Takeaways:
+1️⃣ 
+2️⃣ 
+3️⃣ 
+
+Let's continue pushing the boundaries of what's possible in the world of AI and question-answering. Here's to many more innovations! 🚀
+Shout out to @AIMakerspace !
+
+#LangChain #QuestionAnswering #RetrievalAugmented #Innovation #AI #TechMilestone
+
+Feel free to reach out if you're curious or would like to collaborate on similar projects! 🤝🔥
+```
+
+> #### NOTE: PLEASE SHUTDOWN YOUR INSTANCES WHEN YOU HAVE COMPLETED THE ASSIGNMENT TO PREVENT UNESSECARY CHARGES.

app.py

@@ -0,0 +1,163 @@
+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
+"""
+### 1. CREATE TEXT LOADER AND LOAD DOCUMENTS
+### NOTE: PAY ATTENTION TO THE PATH THEY ARE IN. 
+text_loader = TextLoader("data/paul_graham_essays.txt")
+documents = text_loader.load()
+
+### 2. CREATE TEXT SPLITTER AND SPLIT DOCUMENTS
+text_splitter = RecursiveCharacterTextSplitter(
+    chunk_size=1000,
+    chunk_overlap=30,
+    length_function=len,
+    is_separator_regex=False,
+)
+split_documents = text_splitter.split_documents(documents)
+
+### 3. LOAD HUGGINGFACE EMBEDDINGS
+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)
+    
+    ### 4. INDEX FILES
+    ### NOTE: REMEMBER TO BATCH THE DOCUMENTS WITH MAXIMUM BATCH SIZE = 32
+
+hf_retriever = vectorstore.as_retriever()
+
+# -- AUGMENTED -- #
+"""
+1. Define a String Template
+2. Create a Prompt Template from the String Template
+"""
+### 1. DEFINE 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|>
+"""
+
+### 2. CREATE PROMPT TEMPLATE
+rag_prompt = PromptTemplate.from_template(RAG_PROMPT_TEMPLATE)
+
+# -- GENERATION -- #
+"""
+1. Create a HuggingFaceEndpoint for the LLM
+"""
+### 1. CREATE HUGGINGFACE ENDPOINT FOR LLM
+hf_llm = HuggingFaceEndpoint(
+    endpoint_url=HF_LLM_ENDPOINT,
+    max_new_tokens=512,
+    top_k=10,
+    top_p=0.95,
+    typical_p=0.95,
+    temperature=0.01,
+    repetition_penalty=1.03,
+    huggingfacehub_api_token=os.environ["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.
+    """
+
+    ### BUILD LCEL RAG CHAIN THAT ONLY RETURNS TEXT
+    lcel_rag_chain = {"context": itemgetter("query") | hf_retriever, "query": itemgetter("query")}| 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 @@
+# This is Stock-Analyzer App in Beta

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

solution_app.py

@@ -0,0 +1,155 @@
+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 = (
+        {"context": itemgetter("query") | hf_retriever, "query": itemgetter("query")}
+        | 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="")
+
+    for chunk in await cl.make_async(lcel_rag_chain.stream)(
+        {"query": message.content},
+        config=RunnableConfig(callbacks=[cl.LangchainCallbackHandler()]),
+    ):
+        await msg.stream_token(chunk)
+
+    await msg.send()