deepseek-ai/DeepGEMM

Fork 0

mirror of https://github.com/deepseek-ai/DeepGEMM.git synced 2026-11-05 00:47:26 +00:00

[GH-ISSUE #194] Wheel is not supported in the platform #110

New issue

Open

opened 2026-10-26 21:12:49 +00:00 by bsakaguchi · 1 comment

bsakaguchi commented

2026-10-26 21:12:49 +00:00

Owner

Originally created by @acphile on GitHub (Sep 16, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepGEMM/issues/194

Hi, I am trying to install this package through install.sh on 79f48ee15a. However, I get:

ERROR: deep_gemm-2.0.0+79f48ee-cp310-cp310-linux_x86_64.whl is not a supported wheel on this platform.

I am using CUDA Version: 12.8 on H100 device. Did anyone meet this error before?

Originally created by @acphile on GitHub (Sep 16, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepGEMM/issues/194 Hi, I am trying to install this package through `install.sh` on 79f48ee15a82dd5fad5cd9beaa393c1f755e6b55. However, I get: ``` ERROR: deep_gemm-2.0.0+79f48ee-cp310-cp310-linux_x86_64.whl is not a supported wheel on this platform. ``` I am using CUDA Version: 12.8 on H100 device. Did anyone meet this error before?

bsakaguchi commented

2026-10-26 21:12:51 +00:00

Author

Owner

@Lumb3 commented on GitHub (Mar 12, 2026):

The error indicates that the pre-built wheel deep_gemm-2.0.0+79f48ee-cp310-cp310-linux_x86_64.whl is not compatible with your current Python/platform environment. This often happens when:

You are not using Python 3.10 (the wheel is built for cp310).
Your pip version is outdated or your platform’s manylinux tags do not match those used to build the wheel (e.g., glibc version too old).
You are inside a Conda environment with a different platform tag (e.g., linux-64 vs manylinux).

Let’s troubleshoot step by step.

1. Verify your Python version

python --version

Must be Python 3.10.x. If not, switch to Python 3.10 (e.g., via conda create -n deepgemm python=3.10 or pyenv).

2. Check your pip version and supported tags

pip --version                 # should be recent (≥21.x)
pip debug --verbose           # shows all supported wheel tags

Look for tags like cp310-cp310-manylinux_*_x86_64. Compare them with the wheel’s filename. If your system lacks a required manylinux tag, you’ll need to build from source.

3. Check glibc version

ldd --version | head -n1

Manylinux2014 requires glibc ≥ 2.17, manylinux_2_28 requires ≥ 2.28. If your glibc is older than what the wheel expects, you’ll get this error.

4. Build from source (recommended if the wheel is incompatible)

First, ensure you have CUDA 12.8 and PyTorch installed (with matching CUDA). Then, inside the repository root:

pip install .                 # builds from source

or, if you prefer a wheel:

pip install build
python -m build --wheel
pip install dist/deep_gemm-*.whl

5. If using Conda

Conda environments sometimes alter platform tags. Try:

conda install pip            # ensure conda’s pip is used
pip install .                # still fails? try with --no-build-isolation

6. Force installation (I would not recommended)

As a last resort, you can bypass tag checks with --only-binary :all: --find-links . --no-index, but this may lead to runtime errors if the wheel truly is incompatible.

Let me know which step reveals the mismatch, and we can dig deeper!

@Lumb3 commented on GitHub (Mar 12, 2026): The error indicates that the pre-built wheel `deep_gemm-2.0.0+79f48ee-cp310-cp310-linux_x86_64.whl` is not compatible with your current Python/platform environment. This often happens when: - You are not using Python 3.10 (the wheel is built for `cp310`). - Your `pip` version is outdated or your platform’s `manylinux` tags do not match those used to build the wheel (e.g., glibc version too old). - You are inside a Conda environment with a different platform tag (e.g., `linux-64` vs `manylinux`). Let’s troubleshoot step by step. ## 1. Verify your Python version ```bash python --version ``` Must be `Python 3.10.x`. If not, switch to Python 3.10 (e.g., via `conda create -n deepgemm python=3.10` or `pyenv`). ## 2. Check your pip version and supported tags ```bash pip --version # should be recent (≥21.x) pip debug --verbose # shows all supported wheel tags ``` Look for tags like `cp310-cp310-manylinux_*_x86_64`. Compare them with the wheel’s filename. If your system lacks a required `manylinux` tag, you’ll need to build from source. ## 3. Check glibc version ```bash ldd --version | head -n1 ``` Manylinux2014 requires glibc ≥ 2.17, manylinux_2_28 requires ≥ 2.28. If your glibc is older than what the wheel expects, you’ll get this error. ## 4. Build from source (recommended if the wheel is incompatible) First, ensure you have CUDA 12.8 and PyTorch installed (with matching CUDA). Then, inside the repository root: ```bash pip install . # builds from source ``` or, if you prefer a wheel: ```bash pip install build python -m build --wheel pip install dist/deep_gemm-*.whl ``` ## 5. If using Conda Conda environments sometimes alter platform tags. Try: ```bash conda install pip # ensure conda’s pip is used pip install . # still fails? try with --no-build-isolation ``` ## 6. Force installation (I would not recommended) As a last resort, you can bypass tag checks with `--only-binary :all: --find-links . --no-index`, but this may lead to runtime errors if the wheel truly is incompatible. Let me know which step reveals the mismatch, and we can dig deeper!

No labels

pull-request

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

deepseek-ai/DeepGEMM#110

No description provided.

Rows
Columns