Download Latest Version
v2.2.2 source code.zip (64.6 MB)
Get an email when there's a new version of GPUStack
| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| gpustack-2.1.2-py3-none-any.whl | 2026-04-21 | 12.6 MB | |
| README.md | 2026-04-17 | 1.4 kB | |
| v2.1.2 source code.tar.gz | 2026-04-17 | 53.1 MB | |
| v2.1.2 source code.zip | 2026-04-17 | 53.6 MB | |
| Totals: 4 Items | 119.3 MB | 0 | |
Enhancements
- Support configuring image pull authentication at the model deployment level. (Issue [#5084])
- Mark MetaX as stable and dropped the experimental label. (Issue [#5147])
Bug Fixes
- Fixed an issue where the Worker API key record increased unexpectedly in all-in-one deployments. (Issue [#5115])
- Fixed a benchmark throughput test failure prompting trust_remote_code=True even when the model was already configured with that parameter. (Issue [#5126], [#5138])
- Fixed a failure to detect DCU on the Hygon platform due to inability to access the /dev/dri/card1 device. (Issue [#5104])
- Fixed incorrect parsing of backend parameters containing spaces when the PYPI_PACKAGES_INSTALL environment variable was configured. (Issue [#5093])
- Fixed a GPUStack server crash when starting a custom llama.cpp backend with the -hf argument. (Issue [#5090])
- Fixed a compatibility issue with DTK 26.04. (Issue [#5085])
- Fixed an incorrect runtime version check for custom vLLM versions. (Issue [#5076])
- Fixed a failure to detect Hygon devices caused by PCI device ID changes. (Issue [#5070])
- Fixed the Quay.io mirror missing the latest tag. (Issue [#5064])
- Fixed a GPU detection error in Windows WSL2 environments. (Issue [#5047])
- Fixed a startup failure when using the community llama.cpp backend. (Issue [#5015])
- Fixed a failure to retrieve the model pretrained config during deployment. (Issue [#4855])
