search

acids-ircam / RAVE

Official implementation of the RAVE model: a Realtime Audio Variational autoEncoder
Other
1.29k stars 175 forks source link

Cannot allocate memory error with SimpleDataset() in train_rave.py #85

Closed jreus closed 2 years ago

jreus commented 2 years ago

Hey all, I'm trying to train a rave model and am persistently getting a lmdb.MemoryError. Has anyone encountered such a thing in their training attempts?

$ CUDA_VISIBLE_DEVICES=0 python train_rave.py --name blonk_nolatency_1 --wav ~/datasets/unconditional/Blonk/Vocalor/ --preprocessed ../jobs/rave/blonk_nolatency_1/tmp --sr 48000 --data-size 16 --no-latency true

Traceback (most recent call last):
  File "/its/home/user/RAVE/train_rave.py", line 98, in <module>
    dataset = SimpleDataset(
  File "/its/home/user/miniconda3/envs/rave/lib/python3.10/site-packages/udls/simple_dataset.py", line 66, in __init__
    self.env = SimpleLMDBDataset(out_database_location, map_size)
  File "/its/home/user/miniconda3/envs/rave/lib/python3.10/site-packages/udls/base_dataset.py", line 12, in __init__
    self.env = lmdb.open(out_database_location,
lmdb.MemoryError: ../jobs/rave/blonk_nolatency_1/tmp: Cannot allocate memory
caillonantoine commented 2 years ago

Do you have enough disk space ? You can check it using df -h

jreus commented 2 years ago

Do you have enough disk space ? You can check it using df -h

Yes, lots of disk space. But there was some kind of strange incompatibility in my dependencies. I tried reinstalling anaconda from scratch and now it works. Or.. more accurately, I'm now getting a less showstopping error ;-) (see issue 86 )