Training
Wanda
Pruning / sparsityReferencePrune by weight × activation-norm — no backprop, no weight update.
- Relieves
- Capacity
- Targets
- Weights
- Format
- Unstructured / 2:4
- Granularity
- Weight × activation-norm scoring
- Structure
- unstructured
- Lifecycle
- PTQ · no gradients
- Calibration
- Small calibration set
- Compression
- ~2×
- Quality
- ≈SparseGPT with no weight update
- Hardware
- Sparse kernels; 2:4 on NVIDIA
- Runtimes
- Transformers
A worked Spark recipe for this method hasn't been written yet — it lives here as a reference point in the ontology.