build-small-hackathon/figment-finetuned-model-archive

Same weights, converted vs nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16. Same weights stored in another dtype (BF16 → BF16/F32), plus 21 new tensors.

Tensors changed: Embeddings 0/1, Norms 0/43, lm_head 0/1

On the Hub · Downloads on Model Pulse