LlamaIndex RAGPlayground

LlamaIndex/RAGPlayground/.devcontainer/devcontainer.json

@@ -0,0 +1,33 @@
+{
+  "name": "Python 3",
+  // Or use a Dockerfile or Docker Compose file. More info: https://containers.dev/guide/dockerfile
+  "image": "mcr.microsoft.com/devcontainers/python:1-3.11-bullseye",
+  "customizations": {
+    "codespaces": {
+      "openFiles": [
+        "README.md",
+        "rag\app.py"
+      ]
+    },
+    "vscode": {
+      "settings": {},
+      "extensions": [
+        "ms-python.python",
+        "ms-python.vscode-pylance"
+      ]
+    }
+  },
+  "updateContentCommand": "[ -f packages.txt ] && sudo apt update && sudo apt upgrade -y && sudo xargs apt install -y <packages.txt; [ -f requirements.txt ] && pip3 install --user -r requirements.txt; pip3 install --user streamlit; echo '✅ Packages installed and Requirements met'",
+  "postAttachCommand": {
+    "server": "streamlit run rag\app.py --server.enableCORS false --server.enableXsrfProtection false"
+  },
+  "portsAttributes": {
+    "8501": {
+      "label": "Application",
+      "onAutoForward": "openPreview"
+    }
+  },
+  "forwardPorts": [
+    8501
+  ]
+}

LlamaIndex/RAGPlayground/README.md

@@ -0,0 +1,78 @@
+## RAG Playground Using LlamaIndex
+[Demo](https://github.com/abhi2596/rag_demo/assets/80634226/08f6c7c4-65e3-49b4-bfb1-9a5db2cce248)
+
+<video width="640" height="480" autoplay>
+  <source src="../../../../_static/RAG_LlamaIndex.mp4" type="video/mp4">
+Your browser does not support the video tag.
+</video>
+
+## Introduction 
+
+The RAG Playground is an application designed to facilitate question-answering tasks based on uploaded PDF documents. It leverages LLamaIndex for RAG functionalities and utilizes Streamlit for the user interface.
+
+## Key Features
+
+- **PDF Upload:** Easily upload PDF files to the application.
+- **Questioning:** Ask questions about the uploaded PDF documents.
+- **RAG Integration:** Utilize LLamaIndex for RAG capabilities.
+- **Embeddings:** Convert text to embeddings using the BAAI/bge-small-en-v1.5 model.
+- **Reranker:** Reorder search results based on relevance to queries.
+- **Streamlit Optimization:** Enhance performance using `@st.experimental_fragment` and `@st.cache_resource`.
+
+## Project Workflow
+
+1. **PDF Processing:**
+   - Load PDF files and extract text using PDFReader.
+   - Load data into Documents in LLamaIndex.
+2. **Chunking and Conversion:**
+   - Chunk text and convert it into nodes using `VectorStoreIndex.from_documents`.
+   - Convert text to embeddings using the BAAI/bge-small-en-v1.5 model.
+3. **Search Optimization:**
+   - Implement a reranker to reorder search results based on query relevance.
+   - Display top-ranked results after reranking.
+4. **Interface Optimization:**
+   - Build the user interface using Streamlit.
+   - Optimize Streamlit performance with `@st.experimental_fragment` and `@st.cache_resource`.
+
+## Tech Stack Used
+
+- LLamaIndex
+- Streamlit
+- BAAI/bge-small-en-v1.5 model
+
+## Repository and Deployment
+Github - https://github.com/abhi2596/UnifyAI_RAG_playground/tree/main
+Streamlit App - https://unifyai-rag-playground.streamlit.app/
+
+Instructions to run locally:
+
+1. First create a virtual environment in python 
+
+```
+python -m venv <virtual env name>
+```
+2. Activate it and install poetry 
+
+```
+source <virtual env name>/Scripts/activate - Windows
+source <virtual env name>/bin/activate - Linux/Unix
+pip install poetry
+```
+3. Clone the repo
+
+```
+git clone https://github.com/abhi2596/UnifyAI_RAG_playground/tree/main
+```
+4. Run the following commands
+
+```
+poetry install 
+cd rag
+streamlit run app.py
+```
+
+## Contributors
+
+| Name | GitHub Profile |
+|------|----------------|
+| Abhijeet Chintakunta | [abhi2596](https://github.com/abhi2596) |

LlamaIndex/RAGPlayground/pyproject.toml

@@ -0,0 +1,27 @@
+[tool.poetry]
+name = "rag"
+version = "0.1.0"
+description = ""
+authors = ["abhi2596 <abhijeetchintakunta@gmail.com>"]
+readme = "README.md"
+
+[tool.poetry.dependencies]
+python = ">=3.9,<3.9.7 || >3.9.7,<4.0"
+streamlit = "^1.33.0"
+llama-index = "^0.10.30"
+llama-index-core = "^0.10.30"
+llama-index-readers-file = "^0.1.19"
+beautifulsoup4 = "^4.12.3"
+html5lib = "^1.1"
+unifyai = "^0.8.2"
+pypdf2 = "^3.0.1"
+llama-index-llms-openai-like = "^0.1.3"
+fitz = "^0.0.1.dev2"
+pymupdf = "^1.24.2"
+sentence-transformers = "^2.2.2"
+llama-index-embeddings-huggingface = "^0.2.0"
+
+
+[build-system]
+requires = ["poetry-core"]
+build-backend = "poetry.core.masonry.api"

LlamaIndex/RAGPlayground/rag/app.py

@@ -0,0 +1,105 @@
+import streamlit as st 
+from llama_index.core import VectorStoreIndex, Settings,Document
+from llama_index.core.embeddings import resolve_embed_model
+from llama_index.core.postprocessor import SentenceTransformerRerank
+from unify_llm import Unify
+from PyPDF2 import PdfReader
+import unify
+from llama_index.core.node_parser import SentenceSplitter
+
+# load_dotenv()
+
+# add dynamic routing
+# look into caching again and also mp_fragment and load_llm and clear chat history 
+# app should be fast
+
+def reset():
+    st.session_state.messages = []
+
+st.title("Chat with Data")
+
+api_key = st.sidebar.text_input("Unify AI Key",type="password")
+
+def provider(model_name):
+    dynamic = st.toggle("Dynamic Routing")
+    if dynamic:
+        providers = ["lowest-input-cost","lowest-output-cost","lowest-itl","lowest-ttft","highest-tks-per-sec"]
+        provider_name = st.selectbox("Select a Provider",options=providers,index=1)
+    else:
+        provider_name = st.selectbox("Select a Provider",options=unify.list_providers(model_name))
+    return provider_name
+
+@st.experimental_fragment
+def mp_fragment():
+    model_name = st.selectbox("Select Model",options=unify.list_models(),index=7)
+    provider_name = provider(model_name)
+    return model_name,provider_name
+
+def load_models(model_name,provider_name):
+    Settings.llm = Unify(model=f"{model_name}@{provider_name}",api_key=api_key)
+
+@st.cache_resource
+def rerank_model():
+    rerank = SentenceTransformerRerank(
+        model="cross-encoder/ms-marco-MiniLM-L-2-v2", top_n=3
+    )
+    return rerank
+
+with st.sidebar:
+    model_name,provider_name = mp_fragment()
+    uploaded_file = st.file_uploader("Upload a PDF file", accept_multiple_files=False,on_change=reset)
+    st.sidebar.button("Clear Chat History",on_click=reset)
+
+
+load_models(model_name,provider_name)
+
+if uploaded_file is not None:
+    @st.cache_resource
+    def vector_store(uploaded_file):
+        reader = PdfReader(uploaded_file)
+        text_list = []
+        for page in reader.pages:
+            text_list.append(page.extract_text())
+        documents = [Document(text=t) for t in text_list]
+        Settings.embed_model = resolve_embed_model(embed_model="local:BAAI/bge-small-en-v1.5")
+        index = VectorStoreIndex.from_documents(documents)
+        return index 
+
+    index = vector_store(uploaded_file)
+    chat_engine = index.as_chat_engine(chat_mode="context",verbose=True,similarity_top_k=10,node_postprocessors=[rerank_model()])
+
+# Initialize chat history
+if "messages" not in st.session_state:
+    st.session_state.messages = []
+
+# Display chat messages from history on app rerun
+for message in st.session_state.messages:
+    with st.chat_message(message["role"]):
+        st.markdown(message["content"])
+
+if len(st.session_state.messages) == 0:
+    with st.chat_message("assistant"):
+        initial = " Follow this steps before asking questions about your data \n 1. Enter Unify API key \n 2. Select a Model and Provider using the sidebar \n 3. Upload a PDF file which is not encrypted \n 4. Any changes to sidebar will reset chat"
+        st.markdown(initial)
+        st.session_state.messages.append({"role": "assistant", "content": initial})
+
+# Accept user input
+if prompt := st.chat_input():
+    # Add user message to chat history
+    if api_key == "" and uploaded_file is None:
+        st.warning("Please Enter a Unify API key and upload a file")
+    elif api_key is None:
+        st.warning("Please Enter a Unify API key")
+    elif uploaded_file is None:
+        st.warning("Upload a File")   
+    else: 
+        st.session_state.messages.append({"role": "user", "content": prompt})
+        # Display user message in chat message container
+        with st.chat_message("user"):
+            st.markdown(prompt)
+        # Display assistant response in chat message container
+        with st.chat_message("assistant"):
+            response = chat_engine.stream_chat(prompt)
+            response = st.write_stream(response.response_gen)
+            # response = st.write(response)
+            st.session_state.messages.append({"role": "assistant", "content": response})

LlamaIndex/RAGPlayground/rag/unify_llm.py

@@ -0,0 +1,26 @@
+from llama_index.llms.openai_like import OpenAILike
+import os 
+from typing import Any,Optional
+
+class Unify(OpenAILike):
+    def __init__(
+        self,
+        model: str,
+        api_key: Optional[str] = None,
+        api_base: str = "https://api.unify.ai/v0",
+        is_chat_model: bool = True,
+        **kwargs: Any,
+    ) -> None:
+        api_key = api_key or os.environ.get("UNIFY_API_KEY", None)
+        super().__init__(
+            model=model,
+            api_key=api_key,
+            api_base=api_base,
+            is_chat_model=is_chat_model,
+            **kwargs,
+        )
+
+    @classmethod
+    def class_name(cls) -> str:
+        """Get class name."""
+        return "UnifyLLM"

LlamaIndex/README.md

@@ -1,5 +1,5 @@
 # LlamaIndex Projects
-This folder contains various projects built using the LangChain Unify Integration. Please headover to the corresponding folder of the project for more details. 
+This folder contains various projects built using the LLamaIndex Unify Integration. Please headover to the corresponding folder of the project for more details. 
 
 ## Introduction
 Provide a brief introduction to your project here. Describe what your project demonstrates, the tech stack used, the motivation behind the project, and briefly explain the necessary concepts used. Feel free to break down this section into multiple subsections depending on your project.