Add support for GPUs with cuda compute capability 5.0

[wukongzhikong]

The current version is incompatible with GPUs with cuda compute capability 5.0(NVIDIA Maxwell architecture), which will cause a crash when trying to chat with models on these devices. Please update the cuda llama backend to support it.

[wukongzhikong]

One more suggestion: list the supported cuda compute capability versions, as ollama does: https://github.com/ollama/ollama/blob/main/docs/gpu.md.

[cebtenzzre]

@wukongzhikong Could you please try the installer here? I attempted to enable compute 5.0 support, but I do not have a compute 5.0 GPU to test on.

[cebtenzzre]

@TheTinfoilTricorn a shot in the dark I know, but I saw you had a GTX 750 Ti. would you happen to be able to test the above build?

[cebtenzzre]

cc @homeworker @s5s00000 @kbvtjkjay

[homeworker]

Work for me, nvidia geforce gtx 950m. Thanks

[wukongzhikong]

@wukongzhikong Could you please try the installer here? I attempted to enable compute 5.0 support, but I do not have a compute 5.0 GPU to test on.

I just tested the new installer, works for now for my gtx 960m. Thank you!