Closed YuhengHuang42 closed 2 years ago
Hello!
Thank you for the report. What happens when you use os_cache = True
does it also fail ?
Could you give a reduced version of your dataset that I could play with ?
One way to debug these kind of errors is to disable compilation. It will show the origin of the segmentation fault:
from ffcv.compiler import Compiler
Compiler.set_enable(False)
(the code will obviously run much slower)
Closing due to inactivity. Feel free to reopen once you have more information.
+1,I meet same problem
Hi, I'm using FFCV to run my own dataset. The DatasetWriter works just fine, however, I encountered Segmentation fault while using ffcv.loader.
The process is just like this:
And the trace of the error:
I also noticed that the error may be related to
OrderOption
. ForOrderOption. QUASI_RANDOM
the loader will fail after 3 batches of data is returned, while forOrderOption.RANDOM
it will fail after about 50+ batches of data is returned.