Merge pull request #224 from imartinez/feature/sentence-transformers-embeddings

Feature/sentence transformers embeddings

README.md

@@ -1,7 +1,7 @@
 # privateGPT
 Ask questions to your documents without an internet connection, using the power of LLMs. 100% private, no data leaves your execution environment at any point. You can ingest documents and ask questions without an internet connection!
 
-Built with [LangChain](https://github.com/hwchase17/langchain) and [GPT4All](https://github.com/nomic-ai/gpt4all) and [LlamaCpp](https://github.com/ggerganov/llama.cpp)
+Built with [LangChain](https://github.com/hwchase17/langchain), [GPT4All](https://github.com/nomic-ai/gpt4all), [LlamaCpp](https://github.com/ggerganov/llama.cpp), [Chroma](https://www.trychroma.com/) and [SentenceTransformers](https://www.sbert.net/).
 
 <img width="902" alt="demo" src="https://user-images.githubusercontent.com/721666/236942256-985801c9-25b9-48ef-80be-3acbb4575164.png">
 
@@ -12,20 +12,19 @@ In order to set your environment up to run the code here, first install all requ
 pip install -r requirements.txt
 ```
 
-Then, download the 2 models and place them in a directory of your choice.
+Then, download the LLM model and place it in a directory of your choice:
 - LLM: default to [ggml-gpt4all-j-v1.3-groovy.bin](https://gpt4all.io/models/ggml-gpt4all-j-v1.3-groovy.bin). If you prefer a different GPT4All-J compatible model, just download it and reference it in your `.env` file.
-- Embedding: default to [ggml-model-q4_0.bin](https://huggingface.co/Pi3141/alpaca-native-7B-ggml/resolve/397e872bf4c83f4c642317a5bf65ce84a105786e/ggml-model-q4_0.bin). If you prefer a different compatible Embeddings model, just download it and reference it in your `.env` file.
 
 Rename `example.env` to `.env` and edit the variables appropriately.
 ```
 MODEL_TYPE: supports LlamaCpp or GPT4All
 PERSIST_DIRECTORY: is the folder you want your vectorstore in
-LLAMA_EMBEDDINGS_MODEL: (absolute) Path to your LlamaCpp supported embeddings model
 MODEL_PATH: Path to your GPT4All or LlamaCpp supported LLM
-MODEL_N_CTX: Maximum token limit for both embeddings and LLM models
+MODEL_N_CTX: Maximum token limit for the LLM model
+EMBEDDINGS_MODEL_NAME: SentenceTransformers embeddings model name (see https://www.sbert.net/docs/pretrained_models.html)
 ```
 
-Note: because of the way `langchain` loads the `LLAMA` embeddings, you need to specify the absolute path of your embeddings model binary. This means it will not work if you use a home directory shortcut (eg. `~/` or `$HOME/`).
+Note: because of the way `langchain` loads the `SentenceTransformers` embeddings, the first time you run the script it will require internet connection to download the embeddings model itself.
 
 ## Test dataset
 This repo uses a [state of the union transcript](https://github.com/imartinez/privateGPT/blob/main/source_documents/state_of_the_union.txt) as an example.
@@ -55,11 +54,11 @@ Run the following command to ingest all the data.
 python ingest.py
 ```
 
-It will create a `db` folder containing the local vectorstore. Will take time, depending on the size of your documents.
+It will create a `db` folder containing the local vectorstore. Will take 20-30 seconds per document, depending on the size of the document.
 You can ingest as many documents as you want, and all will be accumulated in the local embeddings database. 
 If you want to start from an empty database, delete the `db` folder.
 
-Note: during the ingest process no data leaves your local environment. You could ingest without an internet connection.
+Note: during the ingest process no data leaves your local environment. You could ingest without an internet connection, except for the first time you run the ingest script, when the embeddings model is downloaded.
 
 ## Ask questions to your documents, locally!
 In order to ask a question, run a command like:
@@ -83,7 +82,7 @@ Type `exit` to finish the script.
 # How does it work?
 Selecting the right local models and the power of `LangChain` you can run the entire pipeline locally, without any data leaving your environment, and with reasonable performance.
 
-- `ingest.py` uses `LangChain` tools to parse the document and create embeddings locally using `LlamaCppEmbeddings`. It then stores the result in a local vector database using `Chroma` vector store. 
+- `ingest.py` uses `LangChain` tools to parse the document and create embeddings locally using `HuggingFaceEmbeddings` (`SentenceTransformers`). It then stores the result in a local vector database using `Chroma` vector store. 
 - `privateGPT.py` uses a local LLM based on `GPT4All-J` or `LlamaCpp` to understand questions and create answers. The context for the answers is extracted from the local vector store using a similarity search to locate the right piece of context from the docs.
 - `GPT4All-J` wrapper was introduced in LangChain 0.0.162.
 

example.env

@@ -1,5 +1,5 @@
 PERSIST_DIRECTORY=db
-LLAMA_EMBEDDINGS_MODEL=models/ggml-model-q4_0.bin
 MODEL_TYPE=GPT4All
 MODEL_PATH=models/ggml-gpt4all-j-v1.3-groovy.bin
+EMBEDDINGS_MODEL_NAME=all-MiniLM-L6-v2
 MODEL_N_CTX=1000

ingest.py

@@ -19,7 +19,7 @@ from langchain.document_loaders import (
 
 from langchain.text_splitter import RecursiveCharacterTextSplitter
 from langchain.vectorstores import Chroma
-from langchain.embeddings import LlamaCppEmbeddings
+from langchain.embeddings import HuggingFaceEmbeddings
 from langchain.docstore.document import Document
 from constants import CHROMA_SETTINGS
 
@@ -72,22 +72,23 @@ def main():
     # Load environment variables
     persist_directory = os.environ.get('PERSIST_DIRECTORY')
     source_directory = os.environ.get('SOURCE_DIRECTORY', 'source_documents')
-    llama_embeddings_model = os.environ.get('LLAMA_EMBEDDINGS_MODEL')
-    model_n_ctx = os.environ.get('MODEL_N_CTX')
+    embeddings_model_name = os.environ.get('EMBEDDINGS_MODEL_NAME')
 
     # Load documents and split in chunks
     print(f"Loading documents from {source_directory}")
+    chunk_size = 500
+    chunk_overlap = 50
     documents = load_documents(source_directory)
-    text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
+    text_splitter = RecursiveCharacterTextSplitter(chunk_size=chunk_size, chunk_overlap=chunk_overlap)
     texts = text_splitter.split_documents(documents)
     print(f"Loaded {len(documents)} documents from {source_directory}")
-    print(f"Split into {len(texts)} chunks of text (max. 500 tokens each)")
+    print(f"Split into {len(texts)} chunks of text (max. {chunk_size} characters each)")
 
     # Create embeddings
-    llama = LlamaCppEmbeddings(model_path=llama_embeddings_model, n_ctx=model_n_ctx)
+    embeddings = HuggingFaceEmbeddings(model_name=embeddings_model_name)
     
     # Create and store locally vectorstore
-    db = Chroma.from_documents(texts, llama, persist_directory=persist_directory, client_settings=CHROMA_SETTINGS)
+    db = Chroma.from_documents(texts, embeddings, persist_directory=persist_directory, client_settings=CHROMA_SETTINGS)
     db.persist()
     db = None
 

privateGPT.py

@@ -1,6 +1,6 @@
 from dotenv import load_dotenv
 from langchain.chains import RetrievalQA
-from langchain.embeddings import LlamaCppEmbeddings
+from langchain.embeddings import HuggingFaceEmbeddings
 from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
 from langchain.vectorstores import Chroma
 from langchain.llms import GPT4All, LlamaCpp
@@ -8,7 +8,7 @@ import os
 
 load_dotenv()
 
-llama_embeddings_model = os.environ.get("LLAMA_EMBEDDINGS_MODEL")
+embeddings_model_name = os.environ.get("EMBEDDINGS_MODEL_NAME")
 persist_directory = os.environ.get('PERSIST_DIRECTORY')
 
 model_type = os.environ.get('MODEL_TYPE')
@@ -18,8 +18,8 @@ model_n_ctx = os.environ.get('MODEL_N_CTX')
 from constants import CHROMA_SETTINGS
 
 def main():
-    llama = LlamaCppEmbeddings(model_path=llama_embeddings_model, n_ctx=model_n_ctx)
-    db = Chroma(persist_directory=persist_directory, embedding_function=llama, client_settings=CHROMA_SETTINGS)
+    embeddings = HuggingFaceEmbeddings(model_name=embeddings_model_name)
+    db = Chroma(persist_directory=persist_directory, embedding_function=embeddings, client_settings=CHROMA_SETTINGS)
     retriever = db.as_retriever()
     # Prepare the LLM
     callbacks = [StreamingStdOutCallbackHandler()]