List your company

Synonym Substitution Robustness on ELI5 prompts 32-bit payload Llama3.1-8B (test)

70.7Bit Accuracy (10% Substitution)Ours+
50.10855.45460.866.146May 12, 2026

Evaluation results

MethodLinks
Ours+Payload=32-bit, Backbone=LLAMA3.1-8B, Type=Stateful Encoder2026.0570.764.259.156.254.3
MC2MarkPayload=32-bit, Backbone=LLAMA3.1-8B2026.0567.161.957.855.153.6
BiMarkPayload=32-bit, Backbone=LLAMA3.1-8B2026.0566.661.65855.653.6
MPACPayload=32-bit, Backbone=LLAMA3.1-8B2026.0561.257.555.553.252.4
StealthInkPayload=32-bit, Backbone=LLAMA3.1-8B2026.0560.957.254.953.752.6
MirrorMarkPayload=32-bit, Backbone=LLAMA3.1-8B2026.0554.452.350.951.150.5
RSBHPayload=32-bit, Backbone=LLAMA3.1-8B2026.0550.950.650.349.950.1