Llama-3
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Llama-3-8B n≈200
0ASR
42
Llama-3-8b-Instruct finetuned variants v1 (test)
51.2TSR
16
LLaMA-3 8B
56.03TTFT (ms)
12
Llama-3 70B
293.82Kernel Latency (µs)
8
Llama-3 8B
152.69Kernel-level latency (µs)
8
Llama-3-8B Translate perturbation, 30 tokens 1.0 (test)
0.13Mean P
6
Llama-3-8B GPT-4o Paraphrase, 150 Tokens
0.26Mean P
6
Llama-3-8B GPT-4o Paraphrase, 30 Tokens
29Mean P
6
Llama-3-8B Swap 50%, 30 Tokens
25Mean P
6
Llama-3-8B-Instruct
1SRF
4
Llama-3-70B successful attacks
1,321Unique Queries Count
3
Llama-3-70B
35.2Average Attack Similarity
3