chat appication

.gitignore

@@ -0,0 +1,165 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/#use-with-ide
+.pdm.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/
+
+
+models
+chroma
+*.ipynb

app.py

@@ -0,0 +1,157 @@
+import datetime
+import os
+import uuid
+
+import fitz
+import streamlit as st
+from langchain_text_splitters import RecursiveCharacterTextSplitter
+from llama_cpp import Llama
+
+from datastore import ChromaStore
+from embeddings import Embedding
+
+#### state
+if "chat_history" not in st.session_state:
+    st.session_state.chat_history = []
+if "document_submitted" not in st.session_state:
+    st.session_state.document_submitted = False
+
+
+def phi3(input: str, relevant_chunks: list):
+    llm = Llama(
+        model_path=os.path.join(
+            os.getcwd(),
+            "models",
+            "Phi-3.1-mini-4k-instruct-Q4_K_M.gguf",
+        ),
+        n_ctx=2000,
+        n_threads=1,  # The number of CPU threads to use,
+        n_gpu_layers=0,  # The number of layers to offload to GPU,
+    )
+
+    prompt = f"""CONTENT: {relevant_chunks}\n\nQUESTION: {input}\n\nFrom the given CONTENT, Please answer the QUESTION."""
+
+    output = llm(
+        f"<|user|>\n{prompt}<|end|>\n<|assistant|>",
+        max_tokens=2000,
+        stop=["<|end|>"],
+        echo=True,
+    )
+
+    cleaned_output = output["choices"][0]["text"].split("<|assistant|>", 1)[-1].strip()
+    return cleaned_output
+
+
+def generate_unique_id():
+    unique_id = uuid.uuid4()
+    current_time = datetime.datetime.now().strftime("%Y%m%d%H%M%S%f")
+    combined_id = f"{unique_id}-{current_time}"
+    return combined_id
+
+
+def add_to_vectorstore(content: str, chunk_size: int = 500, chunk_overlap: int = 20):
+    chromastore = ChromaStore(collection_name="pdf_store")
+
+    # delete if already exist
+    if "pdf_store" in chromastore.list_collections():
+        chromastore.delete("pdf_store")
+        st.toast("Old database cleaned!")
+    collection = chromastore.create()
+    # chunkify content
+    text_splitter = RecursiveCharacterTextSplitter(
+        chunk_size=chunk_size,
+        chunk_overlap=chunk_overlap,
+        length_function=len,
+        is_separator_regex=False,
+    )
+    chunks = text_splitter.split_text(content)
+
+    # generate embeddings and ids
+    embeddings, ids = [], []
+    for i, chunk in enumerate(chunks):
+        embeddings.append(Embedding.encode_text(chunk).tolist())
+        ids.append(generate_unique_id())
+
+    # add to vectorstore
+    chromastore.add(
+        collection=collection,
+        embeddings=embeddings,
+        documents=chunks,
+        ids=ids,
+    )
+
+
+def similarity_search(query: str):
+    chromastore = ChromaStore(collection_name="pdf_store")
+    collection = chromastore.create()
+    query_embedding = Embedding.encode_text(query).tolist()
+    return chromastore.query(collection=collection, query_embedding=query_embedding)
+
+
+def main():
+    st.set_page_config(page_icon="🤖", page_title="Phi 3 RAG", layout="wide")
+    st.markdown(
+        """<h1 style="text-align:center;">Phi 3 RAG</h1>""", unsafe_allow_html=True
+    )
+    st.markdown(
+        """<h3 style="text-align:center;">Conversational RAG application that utilizes local stack, <a href="https://huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF">Phi-3 mini 4k instruct GGUF</a> and <a href="https://docs.trychroma.com/getting-started">ChromaDB</h3>""",
+        unsafe_allow_html=True,
+    )
+    layout = st.columns(2)
+
+    with layout[0]:
+        with st.container(border=True, height=550):
+            uploaded_file = st.file_uploader(
+                label="Upload document to search",
+                type="PDF",
+                accept_multiple_files=False,
+            )
+            submit = st.button("submit")
+
+            chunk_size = st.slider(
+                label="Chunk_size", min_value=100, max_value=2000, step=100
+            )
+            chunk_overlap = st.slider(
+                label="Chunk overlap", min_value=10, max_value=500, step=10
+            )
+            if uploaded_file is not None and submit is not False:
+                # load in vectorstore
+                doc = fitz.open(stream=uploaded_file.read(), filetype="pdf")
+                text = ""
+                for page in doc:
+                    text += page.get_text()
+                doc.close()
+
+                # add to vectorstore
+                add_to_vectorstore(text, chunk_size, chunk_overlap)
+                st.session_state.document_submitted = True
+                st.toast("Document added successfully added to vectorstore", icon="✅")
+
+    # chats
+    with layout[1]:
+        with st.container(border=True, height=550):
+            if st.session_state.document_submitted:
+                user_input = st.chat_input("Ask me!")
+                if user_input is not None:
+                    st.session_state.chat_history.append(
+                        {"role": "user", "content": str(user_input)}
+                    )
+
+                    with st.spinner("Thinking..."):
+                        # find on vector store
+                        relevant_chunks = similarity_search(user_input)
+                        response = phi3(
+                            input=user_input, relevant_chunks=relevant_chunks
+                        )
+                        st.session_state.chat_history.append(
+                            {"role": "assistant", "content": str(response)}
+                        )
+
+                # display messages
+                for message in reversed(st.session_state.chat_history):
+                    with st.chat_message(message["role"]):
+                        st.markdown(message["content"])
+
+
+if __name__ == "__main__":
+    main()

datastore.py

@@ -0,0 +1,107 @@
+__import__("pysqlite3")
+import sys
+
+sys.modules["sqlite3"] = sys.modules.pop("pysqlite3")
+
+
+import uuid
+from collections import defaultdict
+from typing import Any, List
+
+import chromadb
+import numpy as np
+from chromadb import Collection
+
+from embeddings import Embedding
+
+
+class ChromaStore:
+    def __init__(
+        self,
+        collection_name: str,
+        storage_path: str = "./chroma",
+        database: str = "database",
+        metadata: dict = {"hnsw:space": "cosine"},
+    ) -> None:
+        """Initiate Chromadb
+        - collection_name(str): name of the collection
+        - metadata(dict): available options for 'hnsw:space' are 'l2', 'ip' or 'cosine'.
+        """
+
+        self.collection_name = collection_name
+        self.metadata = metadata
+        self.storage_path = storage_path
+        self.database = database
+
+        self.client = chromadb.PersistentClient(path=self.storage_path)
+
+    def _health_check(self) -> bool:
+        return isinstance(self.client.heartbeat(), int)
+
+    def create(self):
+        collection = self.client.get_or_create_collection(
+            name=self.collection_name,
+        )
+        return collection
+
+    def add(
+        self,
+        collection: Collection,
+        embeddings: List[float],
+        documents: List[str],
+        ids: List[str],
+    ):
+        """Add embeddings, documents to index or collection.
+
+        Args:
+        - collection: created collection.
+        - embeddings: list of embeddings
+        - documents: text documents
+        - ids: list of ids"""
+        try:
+            collection.add(
+                embeddings=embeddings,
+                ids=ids,
+                documents=documents,
+            )
+        except Exception as e:
+            raise Exception(f"Failed to add documents to Chroma store. {e}")
+
+    def query(
+        self,
+        collection: Collection,
+        query_embedding: List[float],
+        top_k: int = 3,
+    ) -> list:
+        """Retrieve relevant images from chroma database.
+
+        Args:
+        - collection: created collection.
+        - query_embedding: query image embedding.
+        - top_k (int): top k images to retrieve.
+
+        Returns:
+        - list of images along with their score.
+        """
+        result = collection.query(query_embeddings=query_embedding, n_results=top_k)
+        relevant_chunks = [chunk for chunk in result["documents"][0]]
+        return relevant_chunks
+        # scores = [round(score, 3) for score in result["distances"][0]]
+        # return list(zip(relevant_chunks, scores))
+
+    def delete(self, collection_name: str):
+        try:
+            self.client.delete_collection(collection_name)
+            return True
+        except Exception as e:
+            raise Exception("Failed to delete collection", e)
+
+    def list_collections(self):
+        return self.client.list_collections()
+
+    @staticmethod
+    def collection_info(collection: Collection):
+        info = defaultdict(str)
+        info["count"] = collection.count()
+        info["top_10_items"] = collection.peek()
+        return info

embeddings.py

@@ -0,0 +1,12 @@
+from typing import List
+
+import torch
+from sentence_transformers import SentenceTransformer
+
+
+class Embedding:
+    model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
+
+    @classmethod
+    def encode_text(cls, text: str):
+        return cls.model.encode(text)