[1/N] Enable intel GPU for unsloth

[leizhenyuan]

Hi unsloth, we are going to support unsloth intel GPU with several prs.
For the first step we are aiming to support several models with LoRA, and increase our feature in the future (including BNB, FlashAttention, xformers), below is our plan for the first step:

1. add intel dependent packages for PyTorch 2.6 in pyproject.toml
2. generalize device types and refactor device-bias code in init.py
3. refactor device-bias code in utils
4. refactor device-bias code in kernels
5. refactor device-bias code for unsloth-zoo
6. refactor device-bias code for models

Currently, we would like to start on Linux with torch2.6.
In order to reduce the pressure of your review, we strive to ensure the simplicity of each PR.
For this PR, we change pyproject.toml to download intel GPU required packages.

We have tested our series prs on Llama-3.2, DeepSeek-R1-Distill-Qwen and Qwen2, below are the log from Llama-3.2 LoRA BF16

[Sweaterdog]

Does this support Intel CPU fine tuning as well?

[leizhenyuan]

Does this support Intel CPU fine tuning as well?

This wheel support the cpu functionality for PyTorch, but this pr is not aiming to support CPU fine tuning, since finetune LLM is still to heavy for cpu device

[Sweaterdog]

I figured it is too much, but good to know!

[danielhanchen]

Ok this looks good - sorry on the delay!

[danielhanchen]

Also would it be possible to edit https://github.com/unslothai/unsloth/blob/main/unsloth/_auto_install.py thanks :)

[zejun-chen]

Also would it be possible to edit https://github.com/unslothai/unsloth/blob/main/unsloth/_auto_install.py thanks :)

yes. It makes sense. @leizhenyuan In future, we can add the xpu into the installation file here.