Llama 2
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Llama 2 70B v1.0 (inference)
23.5Throughput (TOK/s)
5
Llama-2 7B chat hf
4.73PLS
4
Llama-2 variants ProFLingo evaluation suite
100Form Success Rate (Llama-2)
4
Llama-2 1.3B (train)
0Number of Spikes
4
LLaMA-2-7B (test)
1,495Total Time
4
LLaMA-2 70B chatbot workload
45.2TTFT (ms)
4
Llama-2-70B
0.216Per-step Decoding Latency
4
Llama-2-13B-chat Seen Languages (tl2en)
32.6BLEU
3
Llama-2-13B-chat Seen Languages (en2tl)
18.3BLEU
3
Llama-2-13B-chat Seen Languages (sq2en)
17.5BLEU
3
Llama-2-13B-chat Seen Languages (en2sq)
9.7BLEU
3
Llama-2-13B-chat Seen Languages sk2en
35.4BLEU
3
Llama-2-13B-chat Seen Languages (en2sk)
14.4BLEU
3
Llama-2-13B-chat Seen Languages (mk2en)
34.2BLEU
3
Llama-2-13B-chat Seen Languages (en2mk)
15BLEU
3
Llama-2-13B-chat Seen Languages (en2lv)
8.8BLEU
3
Llama-2-13B-chat Seen Languages (lt2en)
17BLEU
3
Llama-2-13B-chat Seen Languages (en2lt)
7.7BLEU Score
3
Llama-2-13B-chat Seen Languages et2en (test)
18.3BLEU
3
LLaMa-2 1B (val)
2.421Validation Loss
2
LLaMA-2 13B (val)
17.85Perplexity
2
LLaMA 2 chatbot workload 70B
315GPU Power (W)
2
Llama-2 Prompt with Random Search 7B-Chat
91Detection Accuracy
1
Llama-2-13B-chat Seen Languages (tl2en) (test)
—BLEU
0
Llama-2-13B-chat Seen Languages (en2tl) (test)
—BLEU
0