Extreme Compression: 1.58-bit Quantization Shrinks Vision Transformer to 668KB for Browser
Any_Tie_1861 · reddit · 2026-07-31
A developer applied BitNet-style ternary quantization (1.58-bit), typically used for LLMs, to a vision transformer (Swin2SR lightweight super-resolution model). With every weight quantized to -1, 0, or +1, the entire model is compressed to just 668 KB gzipped—smaller than most images it upscales.
The model runs entirely client-side via ONNX Runtime Web with zero data uploads, and is also available in safetensors format. On the Set5 ×2 dataset, it achieves an RGB PSNR of 34.44 dB, a 2.66 dB improvement over bicubic interpolation.
The author honestly notes its limitations: it only supports clean ×2 upscaling, won't rescue heavily JPEG'd images, is non-generative (sharpens existing detail rather than inventing it), and trades some peak fidelity for the drastically smaller footprint.
More from Multimodal
- DeepSeek-V4-Flash vs Gemini 3.6 Flash: Multimodal 3D Generation Showdown — teortaxesTex · 2026-07-31
- ByteDance Launches Seedance 2.5: 30s Generation, API Price Hiked Over 50% — 智东西 · 2026-07-31
- Fan Uses AI Video Tools to Create Live-Action Trailer for Expedition 33 — imjm · 2026-07-31
- Hailuo AI's New Video Model Excels at Slow Motion with Character References — LudovicCreator · 2026-07-31
- Testing MiniMax H3 Motion Control: Rivals Kling 3.0 — OneTrueTreasure · 2026-07-31
- See2Think: Do Multimodal Models Truly Leverage Intermediate Visual States for Reasoning? — Siyu Yan · 2026-07-31