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

Mixed workload

Benchmarks

Task NameDataset NameSOTA ResultTrend
Mixed reasoning tasksMixed workload n = 450
Exact-match Accuracy74.7
21
ML job scheduling and early stoppingMixed workload 50/30/20 LoRA/DPO/RLHF, 32 GPUs, 200 jobs, 5 seeds (E1)
JCT (Time)290
9
Showing 2 of 2 rows