Cesc

@pzelasko I got errors when increase word_size from 1 to 8, can you give me some advice? ``` "2023-08-09T09:46:00+08:00" malloc(): invalid size (unsorted) "2023-08-09T09:46:00+08:00" malloc(): invalid size (unsorted) "2023-08-09T09:46:01+08:00" malloc():...

> > I'm using KaldifeatFbank because it's compatible to Kaldi > > Can you try to use GPU to extract features? > > `KaldifeatFbank` supports GPU. If you are using...

> Can you reduce the number of workers (especially for on the fly features) and see if it helps? Yes I can run at most 8 workers for dataloader and...

Got it, thanks !

@csukuangfj I still got segment fault after exporting this ``` export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python ``` I am using lazy manifest loader: ```python cut_set = CutSet.from_manifests( features=m["features"], supervisions=m["supervisions"], lazy=True, output_path=output_dir / "temp.jsonl.gz", )...

I got following message: ``` Thread 1 "python" received signal SIGSEGV, Segmentation fault. vgetargskeywords (args=0x7fffe051e460, kwargs=0x0, format=0x555555814db5 "O:memoryview", kwlist=0x5555558ad8a0 , p_va=0x7fffff7ff410, flags=0) at /tmp/build/80754af9/python_1599203911753/work/Python/getargs.c:1671 1671 /tmp/build/80754af9/python_1599203911753/work/Python/getargs.c: 没有那个文件或目录. (gdb) bt #0...

@csukuangfj any advice?

@csukuangfj , here is the backtrace from frame 15: ``` #15 0x000055555578c3f1 in _io__Buffered_read1_impl (n=8192, self=0x7fffdb1fc300) at /tmp/build/80754af9/python_1599203911753/work/Modules/_io/bufferedio.c:948 #16 _io__Buffered_read1 (self=0x7fffdb1fc300, args=, nargs=) at /tmp/build/80754af9/python_1599203911753/work/Modules/_io/clinic/bufferedio.c.h:221 #17 0x00005555556c3c55 in method_vectorcall_FASTCALL (func=,...

> The stack traces are all about methods from Python. I cannot find any useful information from them thx anyway

> Does pdb work with seg faults to display python stack traces? Also I recall gdb can be made to display python stack traces but I don’t remember how. thanks...