🚀 Supercharge LM Studio with a Vector Database!
Search images, audio, and other general file types + Text to Speech
Introductory Video • Medium Article
-
🔥 Windows systems - sorry only able to test on Window but feel free to contribute!
-
📁Git
-
🌐Pandoc.
-
Compiler.
Certain dependencies don't have a pre-compiled "wheel" so you must build them with something like Microsoft Build Tools and/or Visual Studio. If you choose Visual Studio, for example, make sure to select the "Desktop development with C++" extension and check the four boxes on the right containing the "SDK."
🪟WINDOWS INSTRUCTIONS
Download the latest "release," extract its contents, and navigate to the "src" folder to run the following commands:
- NOTE: If you clone this repository you WILL NOT get the latest release. Instead, you will development versions of this program which may or may not be stable.
Within the src
folder, open a command prompt and create a virtual environment:
python -m venv .
Activate the virtual environment:
.\Scripts\activate
Run the setup script:
Only for
Windows
for now.
python setup_windows.py
Run this command if you want to doublecheck that you installed the Pytorch and gpu-acceleration software correctly:
python check_gpu.py
🐧LINUX INSTRUCTIONS
Linux users must use Release v3.5.2 until I can update the codebase due to recent major changes. Download the ZIP file for that release and follow the instructions in the readme.md.
🍎APPLE INSTRUCTIONS
MacOS users must use Release v3.5.2 until I can update the codebase due to recent major changes. Download the ZIP file for that release and follow the instructions in the readme.md.
🖥️INSTRUCTIONS🖥
- Every time you want to use the program you must activate the virtual environment first from within the
src
folder.
To start the program run this command:
python gui.py
- Read the User Guide located within the graphical user interface itself.
- Select and download a vector/embedding model from the
Models Tab
.
This program extracts the text from a variety of file types and puts them into the vector database. It also allows you to create summarizes of images and transcriptions of audio files to be put into the database.
In the Create Database
tab, select files you want to add to the database. You can click the Choose Files
button as many times as you want.
The supported file extensions are:
.docx
,.epub
,.txt
,.html
,.enex
,.eml
,.msg
,.csv
,.xls
,.xlsx
,.rtf
,.odt
.
This program uses "vision" models to create summaries of images, which can then be entered into the database and searched. Before inputting images, I highly recommend that you test the various vision models for the one you like the most.
To test a vision model:
- From the
Create Database
tab, select one or more images. - From the
Settings
tab, select the vision model you want to test. - From the
Tools
tab, process the images.
After determining which vision model you like, add images to the database by selecting them from the Create Database
tab like any other file. When you eventually create the database they will be automatically processed.
Supported file types are:
.png
,.jpg
,.jpeg
,.bmp
,.gif
,.tif
,.tiff
Audio files can be transcribed and put into the database to be searched. Before transcribing a long audio file, I highly recommend testing the various Whisper
models on a shorter audio file as well as experimenting with different batch
settings. Your goal should be to use as large of a Whisper
model as your GPU supports and then adjust the batch size to keep the VRAM usage within your available VRAM> to ensure that the settings don't exceed your available VRAM.
Supported audio extensions include, but are not limited to:
.mp3
,.wav
,.m4a
,.ogg
,.wma
To test optimal settings:
- Within the
Tools
tab, select a short audio file. - Select a
Whisper
model. Read more about the size and quantization levels in theUser Guide
. - Process the audio file.
- Within the
Create Database
tab, doubleclick the transcription that was just created. - Skim the
page content
field to get a sense of whether the transcription is accurate enough for your use-case or if you need to selecta more accurateWhisper
model.To avoid getting a shoddy transcription, it is always better IMHO to choose as large a model as possible, reducing the
batch
setting if need be to keep the VRAM requirements within you're available VRAM.
Once you've obtained the optimal settings for your system, it's time to transcribe an audio file into the database:
- Within the
Create Database
tab, delete any transcriptions you don't want entered into the database. - Create new transctiptions you want entered (repeate for multiple files).
Batch processing is not yet available.
- Download a vector model from the
Models
tab. - Within the
Create Database
tab, create the database.
- The
Manage Database
tab allows you to view the contents of all databases that you've created and delete them if you want.
- In the
Query Database
tab, select the database you want to use from the pulldown menu. - Enter your question by typing it or using the
Record Question
button. - Check the
chunks only
checkbox to only receive the relevant contexts. - Click
Submit Question
.- In the
Settings
tab, you can change multiple settings regarding querying the database. More information can be found in the User Guide.
- In the
This program gets relevant chunks from the vector database and forwarding them - along with your question - to LM Studio for an answer!
- Perform the above steps regarding entering a question and choosing settings, but make sure that
Chunks Only
is 🔥UNCHECKED🔥. - Start LM Studio and go to the Server tab on the left.
- Load a model.
- Turn
Apply Prompt Formatting
to "OFF." - On the right side within
Prompt Format
, make sure that all of the following settings are blank:System Message Prefix
System Message Suffix
User Message Prefix
User Message Suffix
- At the top, load a model within LM Studio.
- On the right, adjust the
GPU Offload
setting to your liking. - Within my program, go to the
Settings
tab, select the appropriate prompt format for the model loaded in LM Studio, clickUpdate Settings
. - In LM Studio, click
Start Server.
- In the
Query Database
tab, clickSubmit Question
.
Feel free to report bugs or request enhancements by creating an issue on github or contacting me on the LM Studio Discord server (see below link)!
All suggestions (positive and negative) are welcome. "bbc@chintellalaw.com" or feel free to message me on the LM Studio Discord Server.