Differentiable Model Compression via Pseudo Quantization Noise

Open paper
Benchmark Model Rank Results
image-classification-on-imagenetDIFFQ (λ=1e−2)#563Top 1 Accuracy: 82.0
language-modelling-on-wikitext-103DIFFQ (λ=1, g=16)#29Test perplexity: 18.0

Rank counts only results with a code link.