[Bug] Unsloth fails to import on Intel Arc B580

[abrarfahim-1000]

Unsloth fails to import on Intel Arc B580 because unsloth_zoo/temporary_patches/gpt_oss.py (line 540) calls torch.xpu.memory.mem_get_info(), which is not supported on this GPU.

When I try to run the given snippet on my notebook, the above error occurs. I am running it locally on my own hardware. Due to the error I am unable to use unsloth.

from unsloth import FastLanguageModel
import torch

model_name = "F:/LocalLLM/models/Qwen3-VL-8B-Instruct"

max_seq_length = 2048  # Choose sequence length
dtype = None  # Auto detection

# Load model and tokenizer
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name=model_name,
    max_seq_length=max_seq_length,
    dtype=dtype,
    load_in_4bit=True,
)

[abrarfahim-1000]

I edited the source code to accommodate the torch.xpu.memory.mem_get_info() and changed it to torch.xpu.get_device_properties(0).total_memory, this fixed one issue, but then I ran into another one.

ImportError: cannot import name 'intel' from 'triton._C.libtriton' (f:\miniconda3\envs\ai\Lib\site-packages\triton\_C\libtriton.pyd)

This is the new error which I cannot figure out.

[mmathew23]

@leizhenyuan Could you maybe take a look on your end?

[leizhenyuan]

hi @abrarfahim-1000 glad to hear your feedback :)
First issue torch.xpu.memory.mem_get_info() is not support should be related with driver version and pytorch version, may i know you torch version and where didn't you get this torch?

Second issue should be also a env issue regarding with triton install unproperly for intel xpu.
We recommended torch and triton written within pyproject.toml.
Or you can search from https://download.pytorch.org/whl to found our release pytorch wheel as your will.

[abrarfahim-1000]

Hello, thank you for responding promptly.

I am on pytorch 2.8, as that is the latest which is supported by Intel. They haven't released IPEX for torch 2.9 yet.
for installing torch I use these commands

python -m pip install torch==2.8.0 torchvision==0.23.0 torchaudio==2.8.0 --index-url https://download.pytorch.org/whl/xpu
python -m pip install intel-extension-for-pytorch==2.8.10+xpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/

from Intel® Extension for PyTorch* Installation Guide

then i simply did a

pip install unsloth unsloth-zoo

I am on windows by the way and inside a conda environment,
Can you please tell me the recommended way in which you would want me to install?
Should I use the pyproject.toml to install? As I saw there is trition support under specific versions.
Thank you and apologies for wasting your valuable time.

EDIT:
upon checking I realized unsloth's Intel support is only on Linux. If I understand correctly.