2080ti cuda error

[HuChundong]

[2025-04-05 22:47:06.756] [info] Loading weights from /home/hucd/ComfyUI/models/diffusion_models/svdq-int4-flux.1-dev/transformer_blocks.safetensors
[2025-04-05 22:47:12.273] [warning] Failed to load safetensors using method PRIVATE: CUDA error: invalid argument (at /nunchaku/src/Serialization.cpp:124)
[2025-04-05 22:47:12.810] [warning] Failed to load safetensors using method MIO: CUDA error: invalid argument (at /nunchaku/src/Serialization.cpp:130)
[2025-04-05 22:47:18.489] [warning] Failed to load safetensors using method READ: CUDA error: invalid argument (at /nunchaku/src/Tensor.h:72)
[2025-04-05 22:47:21.866] [warning] Memory not pinned
[2025-04-05 22:47:22.699] [info] Done.

are these errors normal? i can still draw picture.

[lmxyy]

I think it does not matter. @sxtyzhangzk can you double-check this?

[HuChundong]

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.120 Driver Version: 550.120 CUDA Version: 12.4 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA GeForce RTX 2080 Ti Off | 00000000:05:00.0 On | N/A |
| 24% 43C P8 24W / 250W | 201MiB / 22528MiB | 5% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
| 1 NVIDIA GeForce RTX 2080 Ti Off | 00000000:06:00.0 Off | N/A |
| 22% 40C P8 34W / 250W | 9MiB / 22528MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
| 2 NVIDIA GeForce RTX 2080 Ti Off | 00000000:09:00.0 Off | N/A |
| 22% 42C P8 21W / 250W | 9MiB / 22528MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
| 3 NVIDIA GeForce RTX 2080 Ti Off | 00000000:0A:00.0 Off | N/A |
| 33% 40C P8 23W / 250W | 8795MiB / 22528MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Fri_Jan__6_16:45:21_PST_2023
Cuda compilation tools, release 12.0, V12.0.140
Build cuda_12.0.r12.0/compiler.32267302_0

use prebuilt torch 2.6 and cp310 whl

[HuChundong]

2080ti image quality is very poor than 4090.

[Ph0rk0z]

I have issue with invalid symbol for some assert function in torch despite using cuda 12.6 conda, python 11 and torch 2.6. So pre-built wheel is not working for me and I will have to troubleshoot or yet again attempt to build.

But on another project (sage attention), I used the MMA functions from here along with AI created ones. My kernel inferences, but similarly the quality is lower. Like the image is undercooked.

Same hardware, 2080ti 22gb

[liaozeyu20000110]

same question: i can get the result, but not sure if the message is normal

use RTX4090D torch2.6+CUDA12.6

message:
[2025-04-17 10:46:35.376] [warning] Failed to load safetensors using method PRIVATE: CUDA error: invalid argument (at nunchaku/src/Serialization.cpp:124) [2025-04-17 10:46:35.561] [warning] Failed to load safetensors using method MIO: CUDA error: invalid argument (at nunchaku/src/Serialization.cpp:130) [2025-04-17 10:46:36.889] [warning] Failed to load safetensors using method READ: CUDA error: invalid argument (at nunchaku/src/Tensor.h:72) [2025-04-17 10:46:38.565] [warning] Memory not pinned

[lmxyy]

See our FAQ:

Try upgrading your CUDA driver and try setting the environment variable NUNCHAKU_LOAD_METHOD to either READ or READNOPIN.
Some references: #249 #311 #276

[lmxyy]

I'll close this issue for now. If the error persists or you encounter any other problems, feel free to open a new issue.