Open grid-beep opened 1 year ago
I met the same problem.
@xb534 and @grid-beep Thanks for reaching out. It looks like the Python forking is not playing nice with our use of SQLite3. Can you please try again after setting the following env var in your shell.
export MIOPEN_DEBUG_DISABLE_SQL_WAL=1
This will disable the write-ahead logs (WAL) for SQLite, if the issue persists please share additional logs.
@xb534 and @grid-beep Thanks for reaching out. It looks like the Python forking is not playing nice with our use of SQLite3. Can you please try again after setting the following env var in your shell.
export MIOPEN_DEBUG_DISABLE_SQL_WAL=1
This will disable the write-ahead logs (WAL) for SQLite, if the issue persists please share additional logs.
Thank you for your response. By adding export MIOPEN_USER_DB_PATH="xxx" to my shell file, the issue with the "locking protocol" was resolved; however, the explanation behind why this approach works remains unknown to me.
The next time I encounter such a problem and unable to solve it using the methods mentioned above, I will try to use the method you suggested and reply to you here.
@xb534 and @grid-beep Thanks for reaching out. It looks like the Python forking is not playing nice with our use of SQLite3. Can you please try again after setting the following env var in your shell.
export MIOPEN_DEBUG_DISABLE_SQL_WAL=1
This will disable the write-ahead logs (WAL) for SQLite, if the issue persists please share additional logs.
@JehandadKhan Unfortunately, the same problem happened again, the following is the log file.
=========================================================================
I've done everything I can to figure this out, but it's a black box to me, don't know what works, hope this fixes it once and for all. sincere thanks.
The same issue
@JehandadKhan Created internal ticket to resolve this issue. Thanks!
Hi @donglixp @grid-beep @xb534 , could you provide some more information on what you were doing to arrive at the Internal error while accessing SQLite database: locking protocol
error? Does this happen during installation, running one of the Mask2Former demos, training or something else?
When adapting Mask2Former to Pytorch-ROCm
I assume you're following the Mask2Former install steps, but I'm not sure if you're using the Deformable-DETR kernel that is used throughout the Mask2Former demos (like in /Mask2Former/demo$ python3 demo.py
). In particular, the last two install steps include:
cd mask2former/modeling/pixel_decoder/ops
sh make.sh
to build the Deformable-DETR kernel from this repository. Are you converting this CUDA kernel to HIP and compiling it for ROCm hardware like the person in this issue is attempting to do? If so, please provide more information on how you're doing this.
If you can provide a specific set of steps to reproduce this issue, that would help significantly with the investigation.
As a final aside - ROCm maintains a repository of pretrained transformers from HuggingFace, which includes Mask2Former. I'm not sure if this fits your usecase but you may consider trying that instead.
When adapting Mask2Former to Pytorch-ROCm, I am facing a
MIOpen Error: /.../data/driver/MLOpen/src/sqlite_db.cpp:209: Internal error while accessing SQLite database: locking protocol
.Python: 3.8.16 GPU: GFX90 PyTorch is installed by:
pip install torch==1.13.1+rocm5.2 torchvision==0.14.1+rocm5.2 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/rocm5.2
The error logs are as follows:
Logs with
MIOPEN_LOG_LEVEL=6
: proj.log