mlboydaisuke commited on
Commit
3755794
·
verified ·
1 Parent(s): 217a556

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +6 -0
README.md CHANGED
@@ -26,6 +26,12 @@ All variants take and return fp32 tensors — swap the `.pte` file, keep your ap
26
  \*Mac arm64, single process, median of 10 — a reference point for relative cost
27
  only, not a device number (torch eager fp32 on the same machine: 392.2 ms).
28
 
 
 
 
 
 
 
29
  ### Precisions that did not earn a slot
30
 
31
  - **fp16 is not shipped**: worst-output corr 0.986 against fp32 eager, below the 0.995 bar for this precision. The file converts and runs; the numbers do not hold up, so it is left out rather than shipped with a warning.
 
26
  \*Mac arm64, single process, median of 10 — a reference point for relative cost
27
  only, not a device number (torch eager fp32 on the same machine: 392.2 ms).
28
 
29
+ ### Checked in the task's own units
30
+
31
+ Correlation is a first filter. These are the numbers that decide:
32
+
33
+ - **int8** — mask IoU 0.963 median against fp32 (worst 0.927 of five images). The disagreement is boundary pixels, which is what int8 costs on a cutout model.
34
+
35
  ### Precisions that did not earn a slot
36
 
37
  - **fp16 is not shipped**: worst-output corr 0.986 against fp32 eager, below the 0.995 bar for this precision. The file converts and runs; the numbers do not hold up, so it is left out rather than shipped with a warning.