[GH-ISSUE #59] Can't run on RTX 5000 series due to outdated torch etc #44

New issue

Open

opened 2026-10-26 21:10:46 +00:00 by bsakaguchi · 1 comment

bsakaguchi commented

2026-10-26 21:10:46 +00:00

Owner

Originally created by @rubentorresbonet on GitHub (Feb 6, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-VL2/issues/59

To use RTX 5000 series, such as 5080 or 5090, one needs to grab a nightly version of pytorch, e.g. pip install --pre torch --index-url https://download.pytorch.org/whl/nightly/cu128 --force-reinstall.

Otherwise you get errors like

NVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_70 sm_75 sm_80 sm_86 sm_90.

CUDA error: no kernel image is available for execution on the device

Problem is, this repo explicitely requires torch 2.0.1.

Any plans to update this to support newer versions of torch?

If anyone knows of any workaround, please let me know.

Thanks!

Originally created by @rubentorresbonet on GitHub (Feb 6, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-VL2/issues/59 To use RTX 5000 series, such as 5080 or 5090, one needs to grab a nightly version of pytorch, e.g. `pip install --pre torch --index-url https://download.pytorch.org/whl/nightly/cu128 --force-reinstall`. Otherwise you get errors like ``` NVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation. The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_70 sm_75 sm_80 sm_86 sm_90. CUDA error: no kernel image is available for execution on the device ``` Problem is, this repo explicitely requires torch 2.0.1. Any plans to update this to support newer versions of torch? If anyone knows of any workaround, please let me know. Thanks!

bsakaguchi commented

2026-10-26 21:10:47 +00:00

Author

Owner

@rubentorresbonet commented on GitHub (Feb 6, 2025):

Update: it seems I hacked this to work on wsl2. I believe the key was:

pip install -U --pre xformers --index-url https://download.pytorch.org/whl/nightly/cu128
pip install --pre torch --index-url https://download.pytorch.org/whl/nightly/cu128 --force-reinstall

@rubentorresbonet commented on GitHub (Feb 6, 2025): Update: it seems I hacked this to work on wsl2. I believe the key was: ``` pip install -U --pre xformers --index-url https://download.pytorch.org/whl/nightly/cu128 pip install --pre torch --index-url https://download.pytorch.org/whl/nightly/cu128 --force-reinstall ```

bsakaguchi referenced this issue

2026-10-26 21:12:32 +00:00

[PR #44] Fix README: Correct DeepSeek Homepage Image #135