v1.0.3

KeepNPU now uses an FP16 AI Core workload by default so nputop UTL reflects active compute. Vector/ReLU remains available with --workload vector. This release also propagates workload selection through the service, MCP, REST, and dashboard; improves unconditional utilization and shutdown behavior; and validates AI Core memory requirements before startup.

v1.0.2

What's Changed

• Fix Ascend busy-threshold utilization by @pzheng460 in #4

Full Changelog: v1.0.1...v1.0.2

v1.0.1

What's Changed

• Update publishing Actions to Node 24 by @pzheng460 in #2
• Fix NPU selection side effects and startup errors by @pzheng460 in #3

Full Changelog: v1.0.0...v1.0.1

KeepNPU 1.0.0

First production release of KeepNPU.

Highlights:

• KeepGPU-compatible blocking CLI and service workflows with NPU-native names
• Ascend torch_npu multi-device controllers with utilization backoff and safe release
• REST, JSON-RPC, MCP stdio, and packaged dashboard
• Live npu-smi 25.x telemetry support validated on Ascend 910B1 and 910B2
• Commands: keep-npu, keep-npu-mcp-server, and keep-npu-alive