Spaces:
Sleeping
Sleeping
Commit History
-
Update app.py
Verified signature23d7a36 -
Update app.py
Verified signature954abe5 -
Update app.py
Verified signature5146d0c -
Update README.md
Verified signature992a107 -
Update requirements.txt
Verified signatureb16c820 -
Update requirements.txt
Verified signature12a7eec -
VERTEX-1B-SNN demo: Spiking Neural Network chat interface
Verified signatureVERTEX-1B-SNN is a 1-billion parameter decoder-only language model built on Spiking Neural Network (SNN) architecture — a paradigm shift from traditional artificial neural networks toward brain-inspired, energy-efficient computation. Unlike standard transformers that use continuous activations, VERTEX employs discrete spike-based communication where neurons fire binary pulses when membrane potential exceeds threshold, dramatically reducing inference energy consumption. Our Spiking Transformer Layer replaces the standard FFN with a temporal computation loop: membrane potential accumulates input current, fires a spike when threshold is reached, resets partially, and repeats across 12 time steps. We use straight-through estimators to enable gradient flow through the non-differentiable spike function, making end-to-end training possible. Sequential Loading: To fit within 16GB VRAM (RTX 5070 Ti), we load only one teacher at a time, rotating every 400 steps. Each teacher uses 4-bit NF4 qua…b9df201 -
initial commit
Verified signatureb82c186