This GitHub repository hosts an innovative project featuring an LSTM-based embedding GPT-like neural network. This network is designed to fuse diverse data modalities such as images, audio, sensor inputs, and text, creating a holistic and human-like sentient AI system with the ability to comprehend and respond across multiple data formats.
Just open the script inference.py and change the MODEL and DEVICE variable if needed and then run the script.
from inference import Inference
if __name__ == '__main__':
inference = Inference('Model Path Here')
output, hidden = inference.run('MENENIUS:')
print(output)clear && python3 train.py --name="Model Name Here" --dataset="Dataset File or Path here" --contextsize=128All training settings can be set in the TRAIN_SETTINGS.py script.
The main settings you want to pay attention to are TEXT_DATASET and max_iters.
TEXT_DATASET is the file path that contains all of yor utf-8 or ascii text. This could be for example Tiny Shakespeare.
max_iters is how many itterations through the dataset you would like to run. I would reccoment setting this to a variable like 5000 if you have a lower teir GPU or CPU but if you have a high teir GPU I would set it to 25000 or 50000.
If you want to finetune a dataset all you need to do is set FINETUNE to True and set LOAD_FILE to the model you want to finetune.
- Train More Public Models
- Additional networks like GRU
-
Train More Public Models -
Additional networks like GRU -
Use last token as new token in inference -
Add more integrations