bhanc commited on
Commit
13d31e3
·
verified ·
1 Parent(s): e9fbad5

Group backend precision notes under a shared heading

Browse files
Files changed (1) hide show
  1. README.md +1 -1
README.md CHANGED
@@ -35,7 +35,7 @@ opaque RGBA grayscale mask; use its foreground intensity as an alpha mask when
35
  compositing the original image. Fine structures can be missed. FP16 favors
36
  speed and may shift mask boundaries; FP32 preserves the reference more closely.
37
 
38
- ## XNNPACK quantization
39
 
40
  The INT8 variant selectively quantizes four expensive convolutions with static
41
  INT8 activations and per-channel INT8 weights. Sensitive layers and skip
 
35
  compositing the original image. Fine structures can be missed. FP16 favors
36
  speed and may shift mask boundaries; FP32 preserves the reference more closely.
37
 
38
+ ## Precision notes
39
 
40
  The INT8 variant selectively quantizes four expensive convolutions with static
41
  INT8 activations and per-channel INT8 weights. Sensitive layers and skip