Update README.md
Browse files
README.md
CHANGED
|
@@ -31,6 +31,7 @@ library_name: diffusers
|
|
| 31 |
<a href='https://hanlab.mit.edu/projects/svdquant'>[Website]</a> 
|
| 32 |
<a href='https://hanlab.mit.edu/blog/svdquant'>[Blog]</a>
|
| 33 |
</div>
|
|
|
|
| 34 |
|
| 35 |
SVDQuant is a post-training quantization technique for 4-bit weights and activations that well maintains visual fidelity. On 12B FLUX.1-dev, it achieves 3.6× memory reduction compared to the BF16 model. By eliminating CPU offloading, it offers 8.7× speedup over the 16-bit model when on a 16GB laptop 4090 GPU, 3× faster than the NF4 W4A16 baseline. On PixArt-∑, it demonstrates significantly superior visual quality over other W4A4 or even W4A8 baselines. "E2E" means the end-to-end latency including the text encoder and VAE decoder.
|
| 36 |
|
|
|
|
| 31 |
<a href='https://hanlab.mit.edu/projects/svdquant'>[Website]</a> 
|
| 32 |
<a href='https://hanlab.mit.edu/blog/svdquant'>[Blog]</a>
|
| 33 |
</div>
|
| 34 |
+
|
| 35 |
|
| 36 |
SVDQuant is a post-training quantization technique for 4-bit weights and activations that well maintains visual fidelity. On 12B FLUX.1-dev, it achieves 3.6× memory reduction compared to the BF16 model. By eliminating CPU offloading, it offers 8.7× speedup over the 16-bit model when on a 16GB laptop 4090 GPU, 3× faster than the NF4 W4A16 baseline. On PixArt-∑, it demonstrates significantly superior visual quality over other W4A4 or even W4A8 baselines. "E2E" means the end-to-end latency including the text encoder and VAE decoder.
|
| 37 |
|