Categories Companies Research

Loading the SOTA2 catalog…

Explore

Categories Companies Research

Contribute

Join SOTA2 Contact

SOTA2

© 2026 SOTA2

Build in Bellevue, USA

WizWand is now SOTA2 Research

Papers Benchmarks Datasets Tasks

LLM Training research benchmarks · SOTA2 Research

LLM Training

Benchmarks

Dataset NameSOTA MethodSort

Dataset NameSOTA methodMetricTrendResultsLast Updated

Llama 70B Emulation 3.3

9.3Time Reduction

8

Feb 26, 2026

Qwen2.5-7B 256K context (train)

OOMB (Sparse Attn)

1,301.6Throughput (tokens/sec)

7

Feb 26, 2026

Qwen2.5-7B 128K context (train)

OOMB (Sparse Attn)

1,394.16Throughput (tokens/sec)

5

Feb 26, 2026

Qwen2.5-7B 64K context (train)

OOMB (Sparse Attn)

1,560.38Throughput (tokens/sec)

5

Feb 26, 2026

Llama2-70B (64 x H100-8)

7.8Iteration Time (s)

4

Feb 26, 2026

1.4Iteration Time (s)

4

Feb 26, 2026

Llama2-7B tpu-v5p-512

2.5Iteration Time (s)

3

Feb 26, 2026

Qwen-3 30B-A3B (64 x B200-8)

4.1Iteration Time (s)

2

Feb 26, 2026

Qwen-3 30B-A3B (tpu-v5p-1024)

12.86Iteration Time (s)

2

Feb 26, 2026

Llama2 70B (tpu-v5p-1024)

11.6Iteration Time (s)

2

Feb 26, 2026

Llama2 70B 64 x Trainium2-16

11.2Iteration Time (s)

1

Feb 26, 2026

Llama2 7B 64 x Trainium2-16

1.2Iteration Time (s)

1

Feb 26, 2026

Popular datasets

Llama2 Qwen Llama Qwen2.5-7B

Follow for updates

Join SOTA2 Contact