List your company
ResearchFormerly WizWand

Latency Amplification Analysis on RepetitionCurse and normal prompts (test)

2.452TTFT (Attack)Llama-3-8B-Instruct
1.518527.8195114.120520.42149Dec 30, 2025
Updated 8mo ago

Evaluation results

MethodLinks
Llama-3-8B-InstructArchitecture=Dense Models2025.122.4521.9521.16
Llama-4-Scout-17B-16EArchitecture=MoE Models2025.123.8662.0021.514
ChatGPT-4Architecture=Undisclosed Architecture2025.124.4751.7022.003
GPT-OSS-20BArchitecture=MoE Models2025.126.1971.6722.399
Qwen3-14BArchitecture=Dense Models2025.126.6037.3590.635
Qwen3-Next-80B-A3BArchitecture=MoE Models2025.127.3651.9522.192
Qwen3-Next-80B-A3B-ThinkingArchitecture=MoE Models2025.127.6782.3772.129
Qwen3-30B-A3BArchitecture=MoE Models2025.128.082.5692.138
DeepSeek-ChatArchitecture=MoE Models2025.128.4352.4222.439
ChatGPT-3.5-TurboArchitecture=Undisclosed Architecture2025.128.7041.852.971
DeepSeek-R1-TurboArchitecture=MoE Models2025.1210.7621.9424.012
GPT-OSS-120BArchitecture=MoE Models2025.1213.222.9512.341
ChatGPT-4o-MiniArchitecture=Undisclosed Architecture2025.1213.4153.4821.772
Kimi-Linear-InstructArchitecture=MoE Models2025.1214.5952.0664.728
Qwen3-Coder-30B-A3B-InstructArchitecture=MoE Models2025.1217.0673.4543.233
Mixtral-8x7BArchitecture=MoE Models2025.1225.7895.2883.063