State the audit method and the twelve strength-A verdicts
Browse files
README.md
CHANGED
|
@@ -93,3 +93,9 @@ Then place each file in the directory its type requires (see the table at the to
|
|
| 93 |
and the limits of what that measurement can show - is in
|
| 94 |
[Backdoor Audit Verdicts](https://github.com/ussoewwin/Hybrid-Sensitivity-Weighted-Quantization/blob/main/md/backdoor/Backdoor_Audit_Verdicts.md). The audit code, the procedure and the raw records are in the same
|
| 95 |
repository.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 93 |
and the limits of what that measurement can show - is in
|
| 94 |
[Backdoor Audit Verdicts](https://github.com/ussoewwin/Hybrid-Sensitivity-Weighted-Quantization/blob/main/md/backdoor/Backdoor_Audit_Verdicts.md). The audit code, the procedure and the raw records are in the same
|
| 95 |
repository.
|
| 96 |
+
|
| 97 |
+
All twelve files carry a strength-A verdict. The weight level is closed by replaying each pack with the
|
| 98 |
+
instrument that produced it - the builder functions of its own family, run on the artifact's source - so the
|
| 99 |
+
replay agrees with the pack value for value. Every file also has on record the loader route it is published
|
| 100 |
+
for, one forward with its INT8 call count and output finiteness, one render, and the protection set it keeps
|
| 101 |
+
in original precision.
|