metadata

license: mit
task_categories:
  - other
language:
  - en
tags:
  - quantum
  - neural-networks
  - belief-states
  - transformers
  - rnn
size_categories:
  - 1B<n<10B

Epsilon-Transformers Belief Analysis Dataset

This dataset contains trained neural network models and their corresponding belief state regression analysis from the Epsilon-Transformers project. The models were trained on four different stochastic processes and analyzed for their ability to learn and represent belief states. See https://github.com/adamimos/epsilon-transformers/tree/quantum-public for codebase which generated this data.

Dataset Structure

epsilon-transformers-belief-analysis/
├── README.md
├── models/          # Model checkpoints and configurations from S3
│   ├── {sweep_id}_{run_id}/
│   │   ├── 0.pt                    # Initial checkpoint
│   │   ├── {final}.pt              # Final checkpoint
│   │   ├── run_config.yaml         # Training configuration
│   │   └── loss.csv                # Training loss data
│   └── ...
└── analysis/        # Belief state regression analysis results
    ├── {sweep_id}_{run_id}/
    │   ├── checkpoint_0.joblib              # Initial checkpoint analysis
    │   ├── checkpoint_{final}.joblib        # Final checkpoint analysis
    │   ├── ground_truth_data.joblib         # Neural network ground truth
    │   ├── markov3_checkpoint_*.joblib      # Classical Markov comparisons
    │   └── markov3_ground_truth_data.joblib # Classical ground truth
    └── ...

Model Mappings

Sweep ID	Run ID	Architecture	Process	Description
20241121152808	48	LSTM	Moon Process	LSTM trained on Moon Process
20241121152808	49	LSTM	Bloch Walk	LSTM trained on Bloch Walk
20241121152808	53	LSTM	FRDN	LSTM trained on FRDN
20241121152808	55	LSTM	Mess3	LSTM trained on Mess3
20241121152808	56	GRU	Moon Process	GRU trained on Moon Process
20241121152808	57	GRU	Bloch Walk	GRU trained on Bloch Walk
20241121152808	61	GRU	FRDN	GRU trained on FRDN
20241121152808	63	GRU	Mess3	GRU trained on Mess3
20241121152808	64	RNN	Moon Process	RNN trained on Moon Process
20241121152808	65	RNN	Bloch Walk	RNN trained on Bloch Walk
20241121152808	69	RNN	FRDN	RNN trained on FRDN
20241121152808	71	RNN	Mess3	RNN trained on Mess3
20241205175736	17	Transformer	Bloch Walk	Transformer trained on Bloch Walk
20241205175736	23	Transformer	Mess3	Transformer trained on Mess3
20250421221507	0	Transformer	Moon Process	Transformer trained on Moon Process
20250422023003	1	Transformer	FRDN	Transformer trained on FRDN

File Formats

Model Files (.pt)

Transformerlens (for transformers) or Pytorch (for RNNs) model checkpoints containing trained model weights and optimizer states.

Analysis Files (.joblib)

Joblib-serialized files containing:

checkpoint_*.joblib: Regression analysis results mapping activations to belief states
ground_truth_data.joblib: True belief states and probabilities for the neural network data
markov3_*.joblib: Classical Markov model comparisons and baselines