Llama
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
meta-llama Llama 8B Instruct Long prompt 3.1
12TTFT (s)
3
Llama 3.1 8B Instruct Medium prompt
4.83TTFT (s)
3
meta-llama Llama-3.1-8B-Instruct Short prompt
3.23TTFT (s)
3
Llama-2 13b-chat-hf
0BER
3
Llama-3.3-70B successful attacks
2,240Unique Queries
3
Llama 70B 3.3
0.266Average Attack Similarity
3
Llama-3.3-70B
0.558Attack Success Rate (ASR)
3
Llama-3-70B
45Attack Success Rate (ASR)
3
Llama Q.
100Success Rate
3
LLAMA 1B 3.2
948,036Semantic Efficiency Score (SCE)
3
Llama-3.2-1B circuits
0.627Clarity
3
Llama3-8B Scenario 4: More layers & Pos. (P={-5,...}, L={0,4,...})
46.38Accuracy
3
Llama3.2-3B Scenario 5: ICL prompts
82.8Accuracy
3
Llama3.2-3B Scenario 4: More layers & Pos.
46.38Accuracy
3
Llama3.2-3B Scenario 3: More layers
80.43Accuracy
3
Llama Scenario 1: Diff. Pos. 3.2-3B
74.1Accuracy
3
Llama-2-13B-chat Seen Languages (lv2en)
18.2BLEU
3
Llama-2-13B-chat Seen Languages (en2et)
9.1BLEU Score
3
Llama 8B 3
25Calibration Time (mins)
3
LLaMA-2 7B-CHAT (train)
470.4Total Training Time (s)
3
LLaMA-2-7B
35.8Throughput (tok/s)
2
LLaMA 3 (8B)
283.157Inference Latency (s)
2
Llama 3B 3.2
2.11MSE
2
Llama 3.1
48.1Latency (ms)
2
Llama-2 13B-Chat
67Attack Success Rate (ASR)
2