Release b440

Automated release based on commit count b440

What's Changed

• fix(Gemma4Text): dispatch QuantizedKVCache correctly in LLM attention (fixes SharpAI/SwiftLM#71) by @solderzzc in #29

Full Changelog: b439...b440

Release b439

Automated release based on commit count b439

What's Changed

• fix(Gemma4): add pad token (ID=0) to eosTokenIds to prevent infinite … by @solderzzc in #28

Full Changelog: b436...b439

Release b436

Automated release based on commit count b436

What's Changed

• Revert WIP GLM-5.1 commits from main by @solderzzc in #26

Full Changelog: b435...b436

Release b435

Automated release based on commit count b435

What's Changed

• fix(Evaluate): flatten prompt in TokenRing.loadPrompt to handle 2-D VLM inputs by @solderzzc in #25

Full Changelog: b434...b435

Release b434

Automated release based on commit count b434

What's Changed

• fix(Gemma4): correct double-transpose bug in K-eq-V attention path (closes ml-explore#59) by @solderzzc in #23

Full Changelog: b432...b434

Release b432

Automated release based on commit count b432

What's Changed

• Fix: LFM2.5 VL projector layer norm sanitization by @solderzzc in #22

Full Changelog: b430...b432

Release b430

Automated release based on commit count b430

What's Changed

• Fix: Gemma 4 MoE Loading Failure (gemma4_text) by @solderzzc in #21

Full Changelog: b426...b430

Release b426

Automated release based on commit count b426

What's Changed

• Fix Xcode DMG builder by removing path dependency by @solderzzc in #20

Full Changelog: b425...b426

Release b425

Automated release based on commit count b425

What's Changed

• build: exclude test_features.patch by @solderzzc in #18

Full Changelog: b415...b425

Release b415

Automated release based on commit count b415

What's Changed

• ci: fix downstream integration - add submodules:recursive and pre-install metallib by @solderzzc in #19

Full Changelog: b413...b415