supermix 29

supermix 29
Local Qwen 2.5 0.5B with the latest cognitive-activation LoRA adapter, benchmark summary, and practical desktop app
Model Snapshot
Token F1
0.661
delta +0.470
Perplexity
1.20
delta -26.00
Avg Gen
9.64s
delta -2.92s
Train Time
31.2h
eval 500 samples

Full Changelog: 28...29