Qwen3
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Qwen3-4B-Base Transfer 8 benchmarks
32.65Math Accuracy
6
Qwen3 32B (train)
545.29Training Throughput (128K Seq Len)
5
Qwen3-1.7B
234Total Tokens Processed
4
Qwen3 4B Instruct 2507
71SRF
4
Qwen3-30B-A3B
0.42Reconfiguration Time (s)
4
Qwen3 30B-A3B
84.1Throughput (tokens/s)
3
Qwen3-0.6B (val)
31.45Validation Perplexity
3
Qwen3-32B inference vLLM
7.641Model Load Time (s)
3
Qwen3-14B vLLM (inference)
3.082Model Load Time (s)
3
Qwen3-1.7B vLLM (inference)
0.54Model Load Time (s)
3
Qwen3-32B (train)
125.9Optimizer State Memory (MB)
2
Qwen3-30B-A3B (train)
111.4Optimizer State Memory (MB)
2