BitsFusion is a new weight quantization method that compresses the UNet of Stable Diffusion v1.5 from 1.72 GB (FP16) to 219 MB (1.99 bits), achieving a 7.9X compression ratio. This method not only reduces the model size significantly but also improves its performance. The paper introduces a mixed-precision strategy to achieve this compression, transitioning from FP32, making diffusion-based image generation models more efficient and capable of synthesizing high-quality content.
[CV] BitsFusion: 1.99 bits Weight Quantization of Diffusion Model https://t.co/2kK7swQuyE - The paper proposes BitsFusion, a quantization framework that compresses the UNet of Stable Diffusion v1.5 from FP32 to 1.99 bits. - It introduces a mixed-precision strategy that⦠https://t.co/ZAmsR7P0hx
BitsFusion is a new weight quantization method that can quantize the UNet from Stable Diffusion v1.5 to 1.99 bits, achieving a model with 7.9X smaller size (1.72GB vs 219MB) π₯ All while exhibiting even better generation quality than the original one. https://t.co/A3yXVmbCMZ https://t.co/VwGO3gw1Qe
BitsFusion compresses the UNet of Stable Diffusion v1.5 (1.72 GB, FP16) into 1.99 bits (219 MB), achieving a 7.9X compression ratio and even better performance. Paper: BitsFusion: 1.99 bits Weight Quantization of Diffusion Model Link: https://t.co/4DOXS2XQLc Project:β¦ https://t.co/XJrNRrEpjh
BitsFusion: 1.99 bits Weight Quantization of Diffusion Model https://t.co/HrxYGKZpSb https://t.co/VnVs6zjdnN
BitsFusion 1.99 bits Weight Quantization of Diffusion Model Diffusion-based image generation models have achieved great success in recent years by showing the capability of synthesizing high-quality content. However, these models contain a huge number of parameters, https://t.co/QT56XPFwyS
BitsFusion: 1.99 bits Weight Quantization of Diffusion Model Compresses the UNet of Stable Diffusion v1.5 (1.72 GB, FP16) into 1.99 bits (219 MB), achieving a 7.9X compression ratio and even better performance. proj: https://t.co/4FXu5FCdjh abs: https://t.co/qKr14iN9nL https://t.co/zoaY8wPta9
Post-Training Quantization (PTQ) is an effective method for reducing memory costs and computational complexity. When quantizing diffusion transformers, applying existing diffusion quantization methods designed for U-Net faces challenges in preserving quality. ViDiT-Q is a⦠https://t.co/HR7xuiPDLS