Skip to content

NickCody/mydream

Repository files navigation

Dreambox

NOTE: This README is a work in progress.

INSTALL PREREQUISITES

Rumple Dreambox runs on MacOS, Windows, and Linux.

NOTE: On Linux, run the custom torch install below before running the scripts/init-venv.sh script.

MACOS

brew install pyenv
pyenv global 3.12.0
brew install python3.12 # 10 and 11 are probably ok
brew install portaudio
brew install cmake
python3 -m pip install jax-metal

Linux

On H200 GPU's, you need to run nvidia-smi and note the CUDA Version:

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.127.08             Driver Version: 550.127.08     CUDA Version: 12.4     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GH200 480GB             On  |   00000000:DD:00.0 Off |                    0 |
| N/A   39C    P0            253W /  700W |    9446MiB /  97871MiB |     14%      Default |
|                                         |                        |             Disabled |
+-----------------------------------------+------------------------+----------------------+

And run the appropriate torch install, or modify scripts/install-torch-cuda. The cu124 below matches nvidia-smi CUDA version 12.4:

Windows

win-scripts has setup scripts.

SPEECH RECOGNITION SETUP

Download model to client/audio/models:

Run

You need to run the client and the server in two separate terminals. Before running each, you need to install the dependencies:

scripts/init-venv.sh

NOTE: To run the GUI, additionally you need to pip install gui-requirements.txt

Then activate the `venv`::

```bash
source scripts/activate-venv.sh

Then you can run both of these, except in separate terminals:

scripts/client.sh
scripts/server.sh [model-name] # optpional model name

Typical config looks like this:

    "xl": {
        "model_name": "stabilityai/stable-diffusion-xl-refiner-1.0",
        "pipeline_class": "AutoPipelineForImage2Image",
        "parameters": {
            "strength": 0.33,                   # 0 incoming image is strong, 1 weak
            "num_inference_steps": 20,          # Lower for speed, higher for quality          
            "guidance_scale": 12.0,             # How strongly to follow prompt
            "width": 640,                       
            "height": 512,
            "negative_prompt": "painting, anime, illustration"
        },
        "scheduler": {
            "type": "EulerAncestralDiscreteScheduler"
        }
    }

Running

On Windows/Mac/Linux, you need to set:

export HF_API_KEY=<your key>
export SAFETENSOR_HOME=<safetensor directory>

NOTES

  • Configs for server-side image generation models are in server/config.json
  • Default prompt is in client/gui/main_window.py, lame but will fix later

keys in bashrc comment out pyqt and

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published