Onus-v1-SSM

A basic State Space Model (SSM) with A, B, C matrices and an embedding layer. This model was trained on titles from the 'SocialGrep/one-million-reddit-confessions' dataset.

Training Status: Training interrupted at Epoch 1, Batch 733 (out of 1 planned epochs).

Model Details

• Type: Basic Recurrent State Space Model
• Matrices: A (state transition), B (input to state), C (state to output logits)
• Vocabulary Size: 69 (character-level)
• Embedding Dimension: 32
• Hidden State Dimension: 64
• Max Sequence Length Trained On: 128

Files

• pytorch_model.bin: PyTorch state_dict of the trained model.
• A.npy, B.npy, C.npy: NumPy arrays of the core SSM matrices.
• embedding_weights.npy: NumPy array of the character embedding layer weights.
• model_config_and_vocab.json: Contains char_to_idx, idx_to_char mappings and model/training parameters.
• README.md: This model card.

Training Data

Trained on up to 200000 titles from SocialGrep/one-million-reddit-confessions. Dataset content can be sensitive. Model outputs may reflect this.