Memory leak during evaluation

[ytoon]

During evaluation, memory usage increases continuously over time, eventually leading to excessive memory consumption or OOM.

[omarrayyann]

Hello @ytoon! Which task is this? Could you please share the command you ran.

[ytoon]

──────────────────────────────────────────────

python3 molmo_spaces/evaluation/eval_main.py \
molmo_spaces.evaluation.configs.evaluation_configs:DreamZeroPolicyEvalConfig \
--benchmark_dir assets/benchmarks/molmospaces-bench-v1/procthor-10k/FrankaPickDroidMiniBench/FrankaPickDroidMiniBench_json_benchmark_20251231 \
--task_horizon_steps 450

[BlGene]

@ytoon we will update the code soon, let's look again after this.

[BlGene]

@ytoon we have updated, does this problem still occur for you?

[ytoon]

I’m in the process of testing this and will share the results later.

[ytoon]

I still observe memory increasing.
Each time a new scene is set up, memory usage goes up again.
It looks like resources from the previous scene may not be fully released.

[omarrayyann]

@ytoon I will look into this today and update you.

[omarrayyann]

@ytoon apologies for the very late follow up. the issue has been solved in main