Popular repositories Loading

1. FS-SSA-LM-100M FS-SSA-LM-100M Public

   Softmax-free spiking self-attention at 94M parameters on FineWeb-Edu: PPL 37.44 vs 34.46 for a dense Transformer.

   Python 1

2. Few_Spikes_SNN Few_Spikes_SNN Public

   Few-spikes spiking neural networks (MLP & CNN) for ring counting, with surrogate-gradient training, sparsity analysis, and magnitude pruning in PyTorch.

   Python

3. Few_Spikes_Transformer Few_Spikes_Transformer Public

   Few-spikes activations inside a transformer, tested on synthetic Cherenkov ring counting and JetNet (gluon vs quark tagging).

   Python

4. FS_Softmax_Free_Attention FS_Softmax_Free_Attention Public

   Softmax-free spiking self-attention with few-spikes (FS) neurons: signed ON/OFF populations and learnable per-channel thresholds make it as stable as a full-precision transformer.

   Python

5. FS-SSA-GPT FS-SSA-GPT Public

   Autoregressive Transformer powered by Softmax-Free Spiking Self-Attention (FS-SSA).

   Python