hi, I use your eval code to eval libero, but GPU utils only 60%-75% ,never go to 100%
Device 0 [NVIDIA L20] PCIe GEN 4@16x RX: 69.00 KiB/s TX: 20.00 KiB/s
GPU 2505MHz MEM 9000MHz TEMP 58°C FAN N/A% POW 265 / 350 W
GPU[||||||||||||||||||||| 72%] MEM[||||||||| 15.990Gi/47.988Gi]
Device 1 [NVIDIA L20] PCIe GEN 4@16x RX: 59.00 KiB/s TX: 12.00 KiB/s
GPU 2505MHz MEM 9000MHz TEMP 61°C FAN N/A% POW 243 / 350 W
GPU[||||||||||||||||||||| 72%] MEM[||||||||| 15.933Gi/47.988Gi]
Device 2 [NVIDIA L20] PCIe GEN 4@16x RX: 255.0 KiB/s TX: 11.00 KiB/s
GPU 2445MHz MEM 9000MHz TEMP 63°C FAN N/A% POW 246 / 350 W
GPU[|||||||||||||||||||| 69%] MEM[||||||||| 16.013Gi/47.988Gi]
with such code
CUDA_VISIBLE_DEVICES=2 \
python experiments/robot/libero/run_libero_eval.py \
--task_suite_name libero_object \
--action_decoder_path /root/autodl-tmp/ckpt/univla_ckpt/univla-latent-action-model/lam-stage-2.ckpt\
--pretrained_checkpoint /root/autodl-tmp/ckpt/univla_ckpt/univla-7b-224-sft-libero/univla-libero-object \
--save_video True \
--seed 7
hi, I use your eval code to eval libero, but GPU utils only 60%-75% ,never go to 100%
with such code