feat: add Qwen3.6 MoE model support - #521

Open

qinyiqun wants to merge 9 commits into

InfiniTensor:mainfrom

Commits on Jul 31, 2026

feat(qwen3.6): support dense and MoE models
qinyiqun
committed
8b6f1d2
fix(graph): support Qwen3.5 hybrid position metadata
qinyiqun
committed
09664dc

Commits on Aug 3, 2026

refactor: share hybrid decoder and MoE blocks
qinyiqun
committed
ebcf0d1
fix(graph): make hybrid metadata config-driven
qinyiqun
committed
082245f
refactor(qwen3.5): share model wrappers and reset caches
qinyiqun
committed
34f76b4
refactor(cache): centralize hybrid cache allocation
qinyiqun
committed
2c91d30
refactor(config): share hybrid model defaults
qinyiqun
committed
cce29c0
refactor(qwen3-next): reuse text causal LM template
qinyiqun
committed
ce8761d

Commits on Aug 4, 2026

refactor(qwen3.5): align shared MoE with model family
qinyiqun
committed
a33beaf