Feedback about Reporting Issues

There is the following issue on this page: https://docs.pytorch.org/docs/2.14/user_guide/torch_compiler/compile/programming_model.reporting_issues.html

Hi, I have a PC with Windows 10, install GTX 1060 as GPU, i5-4460 CPU and 16GB RAM. When I run a MOSS-Transcribe-Diarize github project in my PC, I get error as below "!handles_.at(i) INTERNAL ASSERT FAILED at "C:\\actions-runner\\_work\\pytorch\\pytorch\\c10\\cuda\\CUDACachingAllocator.cpp":487, please report a bug to PyTorch."
What my torch version is 2.14.0+cu126. And due to the GTX VRAM is less, I take Gemini suggestion run command to start it as below

(venv) D:\AI_pgm>SET PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
(venv) D:\AI_pgm>mtd-subtitle-web --model "D:\\AI_pgm\\MOSS-Transcribe-Diarize\\moss_transcribe_diarize\\OpenMOSS-Team\\snapshots\\20260915"

Please verify what's the issue of I meet. Thanks

Start with the reporting guide at docs.pytorch.org/docs/2.14/user_guide/torch_compiler/compile/programming_model.reporting_issues.html and the reported CUDACachingAllocator.cpp:487 assertion. Reproduce the MOSS-Transcribe-Diarize command on the stated Windows, GTX 1060 setup with torch 2.14.0+cu126 and PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True; done means identifying a supported cause or collecting enough details for PyTorch triage.

Written by the indexing model from the issue text.