nvidia nemotron nano v2 (nemotronh)#15507

Merged

gabe-l-hart merged 20 commits intoggml-org:masterfrom

gabe-l-hart:gabe-l-hart/nvidia-nemotron-nano-15409

Aug 29, 2025

Commits on Aug 25, 2025

fix: Explicitly enable add_bos_token during conversion
gabe-l-hart
committed

Commits on Aug 26, 2025

fix: Use relu2 (LLM_FFN_RELU_SQR) for activation in FFN layers
gabe-l-hart
committed
fix: Only allocate attention cache for attention layers (not non-recurrent)
gabe-l-hart
committed
fix: Move residual add to after every block
gabe-l-hart
committed

Commits on Aug 28, 2025