Skip to content

UMCode 0.4.4

Latest

Choose a tag to compare

@shaktsin shaktsin released this 16 Nov 01:09
· 1 commit to master since this release
Immutable release. Only release title and notes can be modified.

UMCode 0.4.4

ML Workbench

  • Import supported chat models, test their original or locally fine-tuned weights, and promote a tested model to Chat when its tokenizer has a native chat template.
  • Inspect Mixtral, Granite MoE, Qwen2 MoE, and Qwen3 MoE model structures. Small supported models can use local Data, Train, and Inference workflows; larger models remain inspection-only when they exceed execution limits.
  • Fine-tune supported imported causal models on CPU with attention LoRA. Validation selects the best epoch; the original weights remain intact.
  • Ask the Workbench assistant to import, inspect, and prepare data through approval-gated, sandboxed actions, with dataset provenance shown in the UI.
  • Scale selected training work, parameter, and activation limits with available machine resources.

macOS

  • Universal app for Apple Silicon and Intel Macs, signed with Developer ID and notarized by Apple.
  • The optional isolated-compute and Visual QA browser runtimes are not bundled in this DMG.
  • Download UMCode.dmg and verify it against UMCode.dmg.sha256.