Llama
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Llama-3.1-8B 64k sequence length v1
0.05Decoding Latency (s)
5
Llama 3.1
84.17ASR
5
Llama-3-8B (layer 20)
0.986KL Score
4
Llama 8B Instruct 3.1
6.14PLS
4
Llama-3.1 70B non-judge output abstractions
921Number of Proxy Rules
4
Llama-3.1 8B non-judge output abstractions
1,481Count of Proxy Rules
4
LLaMA-2 Pre-training (val)
28.52Perplexity (1.3B tokens)
4
LLaMA 1B (val)
2.434Validation Loss
4
Llama pretraining 124M
0.674Iteration Time (s)
4
Llama 11B-V 3.2
80KMR (alpha)
4
Llama 7b-chat 2
1SRF
4
LLaMA-2 70B
1.62Reconfiguration Time (s)
4
LLaMA-2 13B
0.51Reconfiguration Time (s)
4
LLaMA-2 7B
0.818Reconfiguration Time (s)
4
LLaMA 13B 2
97.81Bit Accuracy
4
LLaMA (seen)
80.68Object Count
4
Llama2-7b (q=32, k=32) (1k)
35.12TFLOPS
4
Llama 8B 8K Shared Context 3.1
24.8TTFT (ms)
4
Llama 8B 4K Shared Context 3.1
24.7TTFT (ms)
4
Llama 8B 1K Shared Context 3.1
21.7TTFT (ms)
4
LLaMA-135M (val)
3.079Validation Loss
4
LLaMA-60M (val)
3.37Validation Loss
4
Llama 8B Pretraining Distribution 3.1
12.3Retain (pp)
4
Llama 70B 3.1
7.3Throughput (tok/s)
4
Llama 1B 3.2
948,036.91SCE (Efficiency)
4