fix(gemma4_moe): re-tie lm_head to active embed_tokens on MoE path - #2601

Merged

yuhezhang-ai merged 4 commits into

NVIDIA-NeMo:mainfrom

Achyuthan-S:Achyuthan-S/fix/gemma4-moe-tied-lm-head

Jun 22, 2026

Commits on Jun 16, 2026

fix(gemma4_moe): re-tie lm_head to active embed_tokens on MoE path
Achyuthan-S
committed
4497260
refactor(gemma4_moe): expose MoE re-tie via tie_weights()
Achyuthan-S
committed
aa6c324
Merge branch 'main' into Achyuthan-S/fix/gemma4-moe-tied-lm-head
Achyuthan-S
authored
b5c96b0

Commits on Jun 17, 2026

fix(gemma4_moe): make lm_head re-tie follow top-level config flag
Achyuthan-S
committed
bed4c65