ByteDance-Seed / ByteDance-Seed/Bagel
训练edit问题
- Dominant language
- Python
- Stars
- 6.2k
- Forks
- 545
- PR merge metrics
- No merged PRs in 30d
Description
/home/pytorch/alinux3/pytorch/aten/src/ATen/native/cuda/IndexKernel.cu:93: operator(): block: [14322,0,0], thread: [123,0,0] Assertion `-sizes[i] <= index && index < sizes[i] && "index out of bounds"` failed.
/home/pytorch/alinux3/pytorch/aten/src/ATen/native/cuda/IndexKernel.cu:93: operator(): block: [14322,0,0], thread: [124,0,0] Assertion `-sizes[i] <= index && index < sizes[i] && "index out of bounds"` failed.
/home/pytorch/alinux3/pytorch/aten/src/ATen/native/cuda/IndexKernel.cu:93: operator(): block: [14322,0,0], thread: [125,0,0] Assertion `-sizes[i] <= index && index < sizes[i] && "index out of bounds"` failed.
/home/pytorch/alinux3/pytorch/aten/src/ATen/native/cuda/IndexKernel.cu:93: operator(): block: [14322,0,0], thread: [126,0,0] Assertion `-sizes[i] <= index && index < sizes[i] && "index out of bounds"` failed.
/home/pytorch/alinux3/pytorch/aten/src/ATen/native/cuda/IndexKernel.cu:93: operator(): block: [14322,0,0], thread: [127,0,0] Assertion `-sizes[i] <= index && index < sizes[i] && "index out of bounds"` failed.
[rank1]:[E606 15:20:10.319952852 ProcessGroupNCCL.cpp:1595] [PG ID 4 PG GUID 10 Rank 1] Process group watchdog thread terminated with exception: CUDA error: device-side assert triggered
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
Exception raised from c10_cuda_check_implementation at /home/pytorch/alinux3/pytorch/c10/cuda/CUDAException.cpp:43 (most recent call first):
frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x96 (0x7f22969961c6 in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libc10.so)
frame #1: c10::detail::torchCheckFail(char const*, char const*, unsigned int, std::string const&) + 0x5c (0x7f2296940360 in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libc10.so)
frame #2: c10::cuda::c10_cuda_check_implementation(int, char const*, char const*, int, bool) + 0x118 (0x7f22979e9898 in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libc10_cuda.so)
frame #3: c10d::ProcessGroupNCCL::WorkNCCL::finishedGPUExecutionInternal() const + 0x56 (0x7f2242834636 in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
frame #4: c10d::ProcessGroupNCCL::WorkNCCL::isCompleted() + 0xa0 (0x7f2242839680 in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
frame #5: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1ba (0x7f224284316a in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
frame #6: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x14d (0x7f2242844e5d in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
frame #7: + 0xd8b74 (0x7f2313660b74 in /lib64/libstdc++.so.6)
frame #8: + 0x93fb (0x7f2314e9e3fb in /lib64/libpthread.so.0)
frame #9: clone + 0x43 (0x7f2314ba2e83 in /lib64/libc.so.6)
terminate called after throwing an instance of 'c10::DistBackendError'
what(): [PG ID 4 PG GUID 10 Rank 1] Process group watchdog thread terminated with exception: CUDA error: device-side assert triggered
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
Exception raised from c10_cuda_check_implementation at /home/pytorch/alinux3/pytorch/c10/cuda/CUDAException.cpp:43 (most recent call first):
frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x96 (0x7f22969961c6 in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libc10.so)
frame #1: c10::detail::torchCheckFail(char const*, char const*, unsigned int, std::string const&) + 0x5c (0x7f2296940360 in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libc10.so)
frame #2: c10::cuda::c10_cuda_check_implementation(int, char const*, char const*, int, bool) + 0x118 (0x7f22979e9898 in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libc10_cuda.so)
frame #3: c10d::ProcessGroupNCCL::WorkNCCL::finishedGPUExecutionInternal() const + 0x56 (0x7f2242834636 in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
frame #4: c10d::ProcessGroupNCCL::WorkNCCL::isCompleted() + 0xa0 (0x7f2242839680 in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
frame #5: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1ba (0x7f224284316a in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
frame #6: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x14d (0x7f2242844e5d in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
frame #7: + 0xd8b74 (0x7f2313660b74 in /lib64/libstdc++.so.6)
frame #8: + 0x93fb (0x7f2314e9e3fb in /lib64/libpthread.so.0)
frame #9: clone + 0x43 (0x7f2314ba2e83 in /lib64/libc.so.6)
Exception raised from ncclCommWatchdog at /home/pytorch/alinux3/pytorch/torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:1601 (most recent call first):
frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x96 (0x7f22969961c6 in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libc10.so)
frame #1: + 0x102288e (0x7f224282288e in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
frame #2: + 0xcd7401 (0x7f22424d7401 in /opt/conda/envs/python3.10.13/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
frame #3: + 0xd8b74 (0x7f2313660b74 in /lib64/libstdc++.so.6)
frame #4: + 0x93fb (0x7f2314e9e3fb in /lib64/libpthread.so.0)
frame #5: clone + 0x43 (0x7f2314ba2e83 in /lib64/libc.so.6)
Contributor guide
No contributing guide indexed for this repository
Research direction
Start with the reported assertion at aten/src/ATen/native/cuda/IndexKernel.cu:93, then inspect the related failure path in torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:1595-1601. Reproduce the training edit failure with CUDA_LAUNCH_BLOCKING=1 as suggested in the report; done means identifying the invalid index source and confirming the training run no longer triggers the device-side assert.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- distributed-systems, machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 20/100