[Bug]: installation from git is no longer possible on none x86 architectures

[loqs]

⚙️ Your current environment

The output of python collect_env.py

### Environment Information ###
Operating System: `Linux-6.14.0-1013-nvidia-aarch64-with-glibc2.39`
Python Version: `3.13.9 (main, Oct 31 2025, 23:01:48) [Clang 21.1.4 ]`
llm-compressor Version: `None`
compressed-tensors Version: `0.12.3a20251114`
transformers Version: `4.57.1`
torch Version: `2.9.1+cu130`
CUDA Devices: `['NVIDIA GB10']`
AMD Devices: `None`

🐛 Describe the bug

63c175b added auto_round to dependencies in setup.py. auto_round depends on tbb which is only available from pypi as built wheels for Windows x86-64 and Linux x86-64 with no source option. This causes building from git to fail on none X86 platforms

🛠️ Steps to reproduce

$ git clone https://github.com/vllm-project/llm-compressor
$ cd llm-compressor/
$ pip install .
Processing /home/loqs/github/llm-compressor
  Installing build dependencies ... done
  Getting requirements to build wheel ... done
  Preparing metadata (pyproject.toml) ... done
Collecting loguru>=0.7.2 (from llmcompressor==0.8.2.dev65+gdb0b68d9f)
  Using cached loguru-0.7.3-py3-none-any.whl.metadata (22 kB)
Collecting pyyaml>=6.0.1 (from llmcompressor==0.8.2.dev65+gdb0b68d9f)
  Using cached pyyaml-6.0.3-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl.metadata (2.4 kB)
Collecting numpy>=2.0.0 (from llmcompressor==0.8.2.dev65+gdb0b68d9f)
  Using cached numpy-2.3.5-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl.metadata (62 kB)
Collecting requests>=2.32.2 (from llmcompressor==0.8.2.dev65+gdb0b68d9f)
  Using cached requests-2.32.5-py3-none-any.whl.metadata (4.9 kB)
Collecting tqdm>=4.66.3 (from llmcompressor==0.8.2.dev65+gdb0b68d9f)
  Using cached tqdm-4.67.1-py3-none-any.whl.metadata (57 kB)
Collecting torch>=2.7.0 (from llmcompressor==0.8.2.dev65+gdb0b68d9f)
  Downloading torch-2.9.1-cp313-cp313-manylinux_2_28_aarch64.whl.metadata (30 kB)
Collecting transformers>=4.54.0 (from llmcompressor==0.8.2.dev65+gdb0b68d9f)
  Using cached transformers-4.57.1-py3-none-any.whl.metadata (43 kB)
Collecting datasets>=4.0.0 (from llmcompressor==0.8.2.dev65+gdb0b68d9f)
  Using cached datasets-4.4.1-py3-none-any.whl.metadata (19 kB)
Collecting accelerate>=1.6.0 (from llmcompressor==0.8.2.dev65+gdb0b68d9f)
  Using cached accelerate-1.12.0-py3-none-any.whl.metadata (19 kB)
Collecting nvidia-ml-py>=12.560.30 (from llmcompressor==0.8.2.dev65+gdb0b68d9f)
  Using cached nvidia_ml_py-13.580.82-py3-none-any.whl.metadata (9.6 kB)
Collecting pillow>=10.4.0 (from llmcompressor==0.8.2.dev65+gdb0b68d9f)
  Using cached pillow-12.0.0-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl.metadata (8.8 kB)
Collecting compressed-tensors>=0.12.3a2 (from llmcompressor==0.8.2.dev65+gdb0b68d9f)
  Using cached compressed_tensors-0.12.3a20251114-py3-none-any.whl.metadata (7.0 kB)
Collecting auto_round@ git+https://github.com/intel/auto-round.git@llmc (from llmcompressor==0.8.2.dev65+gdb0b68d9f)
  Cloning https://github.com/intel/auto-round.git (to revision llmc) to /tmp/pip-install-_w0z5uhp/auto-round_49b3f62efbd4410a9f09d0fae5afc953
  Running command git clone --filter=blob:none --quiet https://github.com/intel/auto-round.git /tmp/pip-install-_w0z5uhp/auto-round_49b3f62efbd4410a9f09d0fae5afc953
  Running command git checkout -b llmc --track origin/llmc
  Switched to a new branch 'llmc'
  branch 'llmc' set up to track 'origin/llmc'.
  Resolved https://github.com/intel/auto-round.git to commit d96d41e742b61bed25a792a1d39ae70e857d47c1
  Installing build dependencies ... done
  Getting requirements to build wheel ... done
  Preparing metadata (pyproject.toml) ... done
Requirement already satisfied: py-cpuinfo in /home/loqs/vllm-venv-3.13/lib/python3.13/site-packages (from auto_round@ git+https://github.com/intel/auto-round.git@llmc->llmcompressor==0.8.2.dev65+gdb0b68d9f) (9.0.0)
Requirement already satisfied: sentencepiece in /home/loqs/vllm-venv-3.13/lib/python3.13/site-packages (from auto_round@ git+https://github.com/intel/auto-round.git@llmc->llmcompressor==0.8.2.dev65+gdb0b68d9f) (0.2.1)
Requirement already satisfied: packaging in /home/loqs/vllm-venv-3.13/lib/python3.13/site-packages (from auto_round@ git+https://github.com/intel/auto-round.git@llmc->llmcompressor==0.8.2.dev65+gdb0b68d9f) (25.0)
INFO: pip is looking at multiple versions of auto-round to determine which version is compatible with other requirements. This could take a while.
ERROR: Ignored the following versions that require a different python version: 1.21.2 Requires-Python >=3.7,<3.11; 1.21.3 Requires-Python >=3.7,<3.11; 1.21.4 Requires-Python >=3.7,<3.11; 1.21.5 Requires-Python >=3.7,<3.11; 1.21.6 Requires-Python >=3.7,<3.11; 1.26.0 Requires-Python >=3.9,<3.13; 1.26.1 Requires-Python >=3.9,<3.13
ERROR: Could not find a version that satisfies the requirement tbb (from auto-round) (from versions: none)
ERROR: No matching distribution found for tbb

[loqs]

@johnnynunez any suggestions? As you have been very helpful with spark issues.

[kylesayrs]

Seems like tbb is a sticking point for some environments.

It might be worth considering making autoround an optional package dependency of LLM Compressor

pip install llm-compressor[autoround]

[yiliu30]

Hi @loqs @kylesayrs the tbb was introduced by auto-round. We have removed it from the integration branch. Please give it a try.
We’re also preparing a patch release (v0.9.1) for #2062.

@hshen14 @thuang6 @chensuyue