feat(cpu): add Qwen3.5 0.8B mobile CPU support - #690

Merged

chenghuaWang merged 14 commits into

UbiquitousLearning:mainfrom

Aharrypotter:perf/qwen35-arm-cpu-gdn-phase3

Jul 29, 2026

Commits on Jul 27, 2026

feat(cpu): add Qwen3.5 hybrid model runner
Aharrypotter
committed
c70a36a
feat(convertor): support host KAI W4A32 packing
Aharrypotter
committed
639ab26
test(cpu): cover Qwen3.5 conversion and state
Aharrypotter
committed
557dda9

Commits on Jul 28, 2026

fix(models): make generation timing request-scoped
Aharrypotter
committed
db8b190
perf(cpu): pre-normalize Qwen3.5 GDN prefill heads
Aharrypotter
committed
0310517
perf(cpu): accelerate contiguous row copies
Aharrypotter
committed
2972a13
perf(cpu): parallelize Qwen3.5 gated delta heads
Aharrypotter
committed
e2492f2
style(models): format Qwen3.5 helper
Aharrypotter
committed
e6fb93d
test(cpu): cover grouped Qwen3.5 GDN parallelism
Aharrypotter
committed
e42ce17
fix(cpu): address Qwen3.5 review findings
Aharrypotter
committed
896bfd4

Commits on Jul 29, 2026