Benchmark RWKV on Long Range Arena Data Prepration Training Commands listops: RWKV_T_MAX=2048 CUDA_VISIBLE_DEVICES=0,5,6,7 RWKV_FLOAT_MODE=fp32 python -m train wandb=null experiment=lra/rwkv-listops trainer.devices=4 cifar: aan: RWKV_FLOAT_MODE=fp16 python -m train trainer.devices=8 experiment=lra/rwkv-aan wandb=null