v2.4.0

2.4.0 (2026-02-12)

Features

• make api/v1 independently importable as a Go sub-module (#571) (2f2c6a3)

v2.3.6

2.3.6 (2026-02-12)

Bug Fixes

• enhance GPU allocation logic and QoS level calculation (#570) (55025d4)

API Module v0.1.1

Usage

go get github.com/NexusGPU/tensor-fusion/api@v0.1.1

Changes

• feat: make api/v1 independently importable as a Go sub-module (#571)
• fix: enhance GPU allocation logic and QoS level calculation (#570)
• feat: refactor vendor specific config to provider mode (#519)
• fix: allocation/resource metrics not set when using percent mode (#544)
• feat: add constants package for shared constants across the project (#537)
• test: unify Ginkgo/Gomega framework, expand test coverage, and add CapacityConfig safety checks (#516)
• feat: gang scheduling (#512)
• BREAKING CHANGE: refactor to heterogeneous device provider mode (#511)
• feat: hypervisor refactor to support heterogenous devices (#508)

v2.3.5

2.3.5 (2026-02-11)

Bug Fixes

• lint (#569) (fb1574f)

v1.57.4

1.57.4 (2026-02-11)

Bug Fixes

• reduce preemption log verbosity to V(4) to avoid log flooding (0f532af)

v1.57.3

1.57.3 (2026-02-11)

Bug Fixes

• enhance GPU allocation logic and QoS level calculation (f4d2279)

v2.3.4

2.3.4 (2026-02-07)

Bug Fixes

• rename VGPU functions to Accel for consistency across platforms (#562) (11640a6)

v2.3.3

2.3.3 (2026-02-05)

Bug Fixes

• env override for single node process, mem limit env var issue (#561) (bd2643e)

v2.3.2

2.3.2 (2026-02-02)

Bug Fixes

• solve preemption issue (#555) (c113d77)

v2.3.1

2.3.1 (2026-02-01)

Bug Fixes

• support log level setting for single_node mode hypervisor, fix single_node mode worker reconcile (#554) (c7d99dc)