COMMITS
August 26, 2026
J
Publish vMLX 1.6.41 updater manifest
Jinho Jang committed
J
Prepare 1.6.41 release head
Jinho Jang committed
J
Fix multimodal latest-turn tool intent
Jinho Jang committed
J
Fix mixed-SWA changed-tail restart reuse
Jinho Jang committed
J
Wait for DSV4 SSD write durability
Jinho Jang committed
J
Bound DSV4 token contract requests
Jinho Jang committed
J
Use absolute proof paths in DSV4 restart gate
Jinho Jang committed
J
Update DSV4 restart cache proof for SSD-only policy
Jinho Jang committed
J
Prevent dead tool directories from reaching new chats
Jinho Jang committed
J
Retain stock MLX as the native MTP verifier default
Jinho Jang committed
J
Enable the native MTP verify kernel on supported M5 runtimes
Jinho Jang committed
J
Test native MTP mutual exclusion through shared policy
Jinho Jang committed
J
Preserve explicit stream interval settings
Jinho Jang committed
J
Revert DSV4 fused-kernel threadgroup selector
Jinho Jang committed
J
Add guarded DSV4 fused-kernel threadgroup selector
Jinho Jang committed
J
Revert default-off DSV4 b3 hybrid decode path
Jinho Jang committed
J
Add default-off DSV4 b3 hybrid decode path
Jinho Jang committed
J
Revert slower DSV4 fused route accumulation experiment
Jinho Jang committed
J
Add guarded DSV4 fused route accumulation experiment
Jinho Jang committed
J
Publish 1.6.40 updater manifest
Jinho Jang committed
J
Prepare 1.6.40 release head
Jinho Jang committed
J
Preserve request axes in SSM companion keys
Jinho Jang committed
J
Align scoped media keys for SSM reuse
Jinho Jang committed
J
Block stale MLX runtime bundles
Jinho Jang committed
J
Add Electron OpenAI token penalty controls
Jinho Jang committed
J
Keep fused Qwen prefill on the one-shot lane
Jinho Jang committed
J
Use fused D256 attention for Qwen prefill tails
Jinho Jang committed
J
Fix MiniMax M3 adaptive stream reasoning boundaries
Jinho Jang committed
J
fix(panel): extend active model startup safely
Jinho Jang committed
J
Show live progress while creating sessions
Jinho Jang committed