Morpheus 1

Implementation of "MORPHEUS-1" from Prophetic AI and "The world’s first multi-modal generative ultrasonic transformer designed to induce and stabilize lucid dreams. "

Installation

pip install morpheus-torch

Usage

• The input is FRMI and EEG tensors.

• FRMI shape is (batch_size, in_channels, D, H, W)

• EEG Embedding is [batch_size, channels, time_samples]

# Importing the torch library
import torch

# Importing the Morpheus model from the morpheus_torch package
from morpheus_torch.model import Morpheus

# Creating an instance of the Morpheus model with specified parameters
model = Morpheus(
    dim=128,  # Dimension of the model
    heads=4,  # Number of attention heads
    depth=2,  # Number of transformer layers
    dim_head=32,  # Dimension of each attention head
    dropout=0.1,  # Dropout rate
    num_channels=32,  # Number of input channels
    conv_channels=32,  # Number of channels in convolutional layers
    kernel_size=3,  # Kernel size for convolutional layers
    in_channels=1,  # Number of input channels for convolutional layers
    out_channels=32,  # Number of output channels for convolutional layers
    stride=1,  # Stride for convolutional layers
    padding=1,  # Padding for convolutional layers
    ff_mult=4,  # Multiplier for feed-forward layer dimension
    scatter = False, # Whether to scatter to 4d representing spatial dimensions
)

# Creating random tensors for input data
frmi = torch.randn(1, 1, 32, 32, 32)  # Random tensor for FRMI data
eeg = torch.randn(1, 32, 128)  # Random tensor for EEG data

# Passing the input data through the model to get the output
output = model(frmi, eeg)

# Printing the shape of the output tensor
print(output.shape)

Code Quality 🧹

• make style to format the code
• make check_code_quality to check code quality (PEP8 basically)
• black .
• ruff . --fix

License

MIT

Todo

• Implement the scatter in the end of the decoder to output spatial outputs which are 4d?

• Implement a full model with the depth of the decoder layers

• Change all the MHAs to Multi Query Attentions

• Double check popular brain scan EEG and FRMI AI papers to double check tensor shape