You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I have recently upgraded my old OneTrainer version to a recent upgraded version from August 2026. I don't know which older version it was that I used before but before the upgrade I was already able to train Flux. 2 [dev] and Flux.2 [klein] 9B base within a reasonable amount of time.
The updating process was successful. I did not change my old settings which previously worked just fine. I have noticed that prior settings from my old version were missing and new once were implemented but I didn't bother.
Unfortunately and to my suprise I neither was able to train with Flux. 2 [dev] nor with Flux.2 [klein] 9B base on the upgraded version of OneTrainer. After starting, the training status remained at 0% for a while and then I got an Error message claiming this:
torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 864.00 MiB. GPU 0 has a total capacity of 15.99 GiB of which 0 bytes is free. Of the allocated memory 44.84 GiB is allocated by PyTorch, and 1.13 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://docs.pytorch.org/docs/stable/notes/cuda.html#optimizing-memory-usage-with-pytorch-cuda-alloc-conf)
Since my hardware was good enough before my GPU cannot be the issue. I found similar issues with the aforementioned error in the discussion thread but the following solution I found did not work for me:
"This helped me: Train Data Type = Float 32 / Gradient CheckPointing = CPU_OFFLOADED / ETA = OFF. In the optimizer: Fused Bask Pass On / Fused Off"
I then switched to the pre-upgrade version of OneTrainer from June 2026. With this prior version Flux.1 [dev] training was working fine as before. Flux.2 [klein] 9B base on the other hand did not work either, stuck at 0% step for a while and then came up with a different error message:
OneTrainer-pre-upgrade\venv\Lib\site-packages\bitsandbytes\backends\cuda\ops.py:464: FutureWarning: _check_is_size will be removed in a future PyTorch release along with guard_size_oblivious. Use _check(i >= 0) instead.
torch._check_is_size(blocksize)
I tried to solve this issue by myself but I have no idea what is going wrong. I would be glad if someone might have a solution what the issue could be.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
I have recently upgraded my old OneTrainer version to a recent upgraded version from August 2026. I don't know which older version it was that I used before but before the upgrade I was already able to train Flux. 2 [dev] and Flux.2 [klein] 9B base within a reasonable amount of time.
The updating process was successful. I did not change my old settings which previously worked just fine. I have noticed that prior settings from my old version were missing and new once were implemented but I didn't bother.
Unfortunately and to my suprise I neither was able to train with Flux. 2 [dev] nor with Flux.2 [klein] 9B base on the upgraded version of OneTrainer. After starting, the training status remained at 0% for a while and then I got an Error message claiming this:
torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 864.00 MiB. GPU 0 has a total capacity of 15.99 GiB of which 0 bytes is free. Of the allocated memory 44.84 GiB is allocated by PyTorch, and 1.13 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://docs.pytorch.org/docs/stable/notes/cuda.html#optimizing-memory-usage-with-pytorch-cuda-alloc-conf)
Since my hardware was good enough before my GPU cannot be the issue. I found similar issues with the aforementioned error in the discussion thread but the following solution I found did not work for me:
"This helped me: Train Data Type = Float 32 / Gradient CheckPointing = CPU_OFFLOADED / ETA = OFF. In the optimizer: Fused Bask Pass On / Fused Off"
I then switched to the pre-upgrade version of OneTrainer from June 2026. With this prior version Flux.1 [dev] training was working fine as before. Flux.2 [klein] 9B base on the other hand did not work either, stuck at 0% step for a while and then came up with a different error message:
OneTrainer-pre-upgrade\venv\Lib\site-packages\bitsandbytes\backends\cuda\ops.py:464: FutureWarning: _check_is_size will be removed in a future PyTorch release along with guard_size_oblivious. Use _check(i >= 0) instead.
torch._check_is_size(blocksize)
I tried to solve this issue by myself but I have no idea what is going wrong. I would be glad if someone might have a solution what the issue could be.
debug_report.log
All reactions