mlboydaisuke commited on
Commit
e0bb772
·
verified ·
1 Parent(s): 9bdd3d4

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +5 -4
README.md CHANGED
@@ -29,10 +29,11 @@ All variants take and return fp32 tensors — swap the `.pte` file, keep your ap
29
 
30
 
31
  The Core ML build is the same graph lowered to Apple's Neural Engine instead of
32
- XNNPACK, which is CPU-only. On an iPhone 17 Pro, Depth-Anything-V2-Small runs
33
- 500.8 ms through XNNPACK and 42.7 ms through Core ML, at half the file size. It
34
- computes in fp16 and is iOS-only; the XNNPACK files stay the portable option and
35
- are what runs on Android.
 
36
 
37
  \*Mac arm64, single process, median of 10 — a reference point for relative cost
38
  only, not a device number (torch eager fp32 on the same machine: 375.5 ms).
 
29
 
30
 
31
  The Core ML build is the same graph lowered to Apple's Neural Engine instead of
32
+ XNNPACK, which is CPU-only. Measured on an iPhone 17 Pro across seven models, it
33
+ runs **3.5x to 13.9x faster (median 12x)** at roughly half the file size — for
34
+ example Depth-Anything-V2-Small at 500.8 ms against 42.7 ms, and MODNet at 81.7 ms
35
+ against 5.9 ms. It computes in fp16 and is iOS-only; the XNNPACK files stay the
36
+ portable option and are what runs on Android.
37
 
38
  \*Mac arm64, single process, median of 10 — a reference point for relative cost
39
  only, not a device number (torch eager fp32 on the same machine: 375.5 ms).