Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Qwen3 family

Benchmarks

Task NameDataset NameSOTA ResultTrend
Training Efficiency ProfilingQwen3 family S=512 (train profiling)
Per-Step Latency (ms)100.2
43
Training Efficiency ProfilingQwen3 family S=1024 (train profiling)
Per-Step Latency (ms)100.6
42
Showing 2 of 2 rows