comfyanonymous
flux ae.safetensors no longer works in new update
Expected Behavior
be able to generate an image
Actual Behavior
Workflow fails when generating an image when loading VAE
Steps to Reproduce
use detaildaemon img2img workflow Flux img2img-DetailDaemon.json
Debug Logs
2024-11-16T03:11:45.971109107Z frame #20: <unknown function> + 0x2b48eff (0x7eb6bfe5eeff in /usr/local/lib/python3.10/dist-packages/torch/lib/libtorch_cpu.so)
2024-11-16T03:11:45.971111117Z frame #21: at::_ops::convolution::call(at::Tensor const&, at::Tensor const&, c10::optional<at::Tensor> const&, c10::ArrayRef<long>, c10::ArrayRef<c10::SymInt>, c10::ArrayRef<long>, bool, c10::ArrayRef<c10::SymInt>, long) + 0x223 (0x7eb6bf603723 in /usr/local/lib/python3.10/dist-packages/torch/lib/libtorch_cpu.so)
2024-11-16T03:11:45.971113267Z frame #22: at::native::conv2d_symint(at::Tensor const&, at::Tensor const&, c10::optional<at::Tensor> const&, c10::ArrayRef<long>, c10::ArrayRef<c10::SymInt>, c10::ArrayRef<long>, long) + 0x1de (0x7eb6bed9daae in /usr/local/lib/python3.10/dist-packages/torch/lib/libtorch_cpu.so)
2024-11-16T03:11:45.971115277Z frame #23: <unknown function> + 0x2d05051 (0x7eb6c001b051 in /usr/local/lib/python3.10/dist-packages/torch/lib/libtorch_cpu.so)
2024-11-16T03:11:45.971117207Z frame #24: at::_ops::conv2d::call(at::Tensor const&, at::Tensor const&, c10::optional<at::Tensor> const&, c10::ArrayRef<long>, c10::ArrayRef<c10::SymInt>, c10::ArrayRef<long>, long) + 0x204 (0x7eb6bfbdb8d4 in /usr/local/lib/python3.10/dist-packages/torch/lib/libtorch_cpu.so)
2024-11-16T03:11:45.971119157Z frame #25: <unknown function> + 0x5f9acc (0x7eb6d4828acc in /usr/local/lib/python3.10/dist-packages/torch/lib/libtorch_python.so)
2024-11-16T03:11:45.971121337Z frame #26: <unknown function> + 0x15fe0e (0x566b5dd5be0e in /usr/bin/python3)
2024-11-16T03:11:45.971123317Z frame #27: _PyObject_MakeTpCall + 0x25b (0x566b5dd525eb in /usr/bin/python3)
2024-11-16T03:11:45.971125257Z frame #28: _PyEval_EvalFrameDefault + 0x6aa1 (0x566b5dd4b1f1 in /usr/bin/python3)
2024-11-16T03:11:45.971127907Z frame #29: <unknown function> + 0x16e4e1 (0x566b5dd6a4e1 in /usr/bin/python3)
2024-11-16T03:11:45.971129847Z frame #30: _PyEval_EvalFrameDefault + 0x6152 (0x566b5dd4a8a2 in /usr/bin/python3)
2024-11-16T03:11:45.971131687Z frame #31: <unknown function> + 0x16e62e (0x566b5dd6a62e in /usr/bin/python3)
2024-11-16T03:11:45.971133566Z frame #32: _PyEval_EvalFrameDefault + 0x2b71 (0x566b5dd472c1 in /usr/bin/python3)
2024-11-16T03:11:45.971135476Z frame #33: <unknown function> + 0x16e62e (0x566b5dd6a62e in /usr/bin/python3)
2024-11-16T03:11:45.971137356Z frame #34: _PyEval_EvalFrameDefault + 0x2b71 (0x566b5dd472c1 in /usr/bin/python3)
2024-11-16T03:11:45.971139406Z frame #35: <unknown function> + 0x16e62e (0x566b5dd6a62e in /usr/bin/python3)
2024-11-16T03:11:45.971141296Z frame #36: _PyEval_EvalFrameDefault + 0x2b71 (0x566b5dd472c1 in /usr/bin/python3)
2024-11-16T03:11:45.971143536Z frame #37: _PyObject_FastCallDictTstate + 0xc4 (0x566b5dd51784 in /usr/bin/python3)
2024-11-16T03:11:45.971145556Z frame #38: _PyObject_Call_Prepend + 0x5c (0x566b5dd6754c in /usr/bin/python3)
2024-11-16T03:11:45.971147506Z frame #39: <unknown function> + 0x2841e0 (0x566b5de801e0 in /usr/bin/python3)
2024-11-16T03:11:45.971149356Z frame #40: _PyObject_MakeTpCall + 0x25b (0x566b5dd525eb in /usr/bin/python3)
2024-11-16T03:11:45.971151376Z frame #41: _PyEval_EvalFrameDefault + 0x6aa1 (0x566b5dd4b1f1 in /usr/bin/python3)
2024-11-16T03:11:45.971154556Z frame #42: <unknown function> + 0x16e62e (0x566b5dd6a62e in /usr/bin/python3)
2024-11-16T03:11:45.971156446Z frame #43: _PyEval_EvalFrameDefault + 0x2b71 (0x566b5dd472c1 in /usr/bin/python3)
Other
No response
I’ll try it, but this is happening on a fresh install. Do I still need to reinstall torchvision and torch audio on a fresh install?
Sent from Gmail Mobile
On Sat, 16 Nov 2024 at 14:18, Dr.Lt.Data @.***> wrote:
Reinstall torch torchvision torchaudio and try again.
— Reply to this email directly, view it on GitHub https://github.com/comfyanonymous/ComfyUI/issues/5630#issuecomment-2480460799, or unsubscribe https://github.com/notifications/unsubscribe-auth/AACMIV3CVJABCRB5QYGQTST2A3WTLAVCNFSM6AAAAABR4LILGOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIOBQGQ3DANZZHE . You are receiving this because you authored the thread.Message ID: @.***>
I’ll try it, but this is happening on a fresh install. Do I still need to reinstall torchvision and torch audio on a fresh install? Sent from Gmail Mobile … On Sat, 16 Nov 2024 at 14:18, Dr.Lt.Data @.> wrote: Reinstall torch torchvision torchaudio and try again. — Reply to this email directly, view it on GitHub <#5630 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AACMIV3CVJABCRB5QYGQTST2A3WTLAVCNFSM6AAAAABR4LILGOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIOBQGQ3DANZZHE . You are receiving this because you authored the thread.Message ID: @.>
How about when using a different model?
First of all, your log indicates an issue in the native code, which is something ComfyUI cannot address or access.
It might be a bug specific to your environment, so it seems necessary to attempt changes, such as switching to a different PyTorch version, in the hope of avoiding the issue.
If not, you may need to suspect a driver or GPU hardware issue.
The issue is occurring regardless of model. It’s happening on a fresh install, after loading vae (ae.safetensors) and vae decode is highlighted.
Tested with flux dev & flux dev gguf q6.
Should I test with schnell or other quantization?
It shouldn’t matter though right? As it’s the vae decoder that’s causing the error.
Sent from Gmail Mobile
On Sat, 16 Nov 2024 at 15:06, Dr.Lt.Data @.***> wrote:
I’ll try it, but this is happening on a fresh install. Do I still need to reinstall torchvision and torch audio on a fresh install? Sent from Gmail Mobile … <#m_5159187043714779168_> On Sat, 16 Nov 2024 at 14:18, Dr.Lt.Data @.> wrote: Reinstall torch torchvision torchaudio and try again. — Reply to this email directly, view it on GitHub <#5630 (comment) https://github.com/comfyanonymous/ComfyUI/issues/5630#issuecomment-2480460799>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AACMIV3CVJABCRB5QYGQTST2A3WTLAVCNFSM6AAAAABR4LILGOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIOBQGQ3DANZZHE https://github.com/notifications/unsubscribe-auth/AACMIV3CVJABCRB5QYGQTST2A3WTLAVCNFSM6AAAAABR4LILGOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIOBQGQ3DANZZHE . You are receiving this because you authored the thread.Message ID: @.>
How about when using a different model?
— Reply to this email directly, view it on GitHub https://github.com/comfyanonymous/ComfyUI/issues/5630#issuecomment-2480472396, or unsubscribe https://github.com/notifications/unsubscribe-auth/AACMIV4D2LAACF5C3ZSSERT2A34INAVCNFSM6AAAAABR4LILGOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIOBQGQ3TEMZZGY . You are receiving this because you authored the thread.Message ID: @.***>
The issue is occurring regardless of model. It’s happening on a fresh install, after loading vae (ae.safetensors) and vae decode is highlighted. Tested with flux dev & flux dev gguf q6. Should I test with schnell or other quantization? It shouldn’t matter though right? As it’s the vae decoder that’s causing the error.
What I meant is not about Flux but about models like SD1.5 and SDXL.
Ah I haven't let me test.
curiously, could it have anthing to do with my docker base being pytorch:2.1.1-py3.10-cuda12.1.1-devel-ubuntu22.04
do I need to be using a more updated base?
That is a very outdated version and not even a stable one. I recommend upgrading PyTorch.
I updated it to pytorch:2.4.0-py3.11-cuda12.4.1-devel-ubuntu22.04
Should be better right?