nvidia/Nemotron-H-8B-Base-8K model repo

Fields

id

sources/hf-eco-nemotron-h-8b-base-8k

URL

https://huggingface.co/nvidia/Nemotron-H-8B-Base-8K

Publisher

NVIDIA

Kind

repo

Published

2025-04-14

Accessed

2026-10-03

notes

config.json (NemotronHForCausalLM, 52 layers, hybrid_override_pattern 'M-M-M-M*-...' with four starred positions) plus model.safetensors.index.json, in which q/k/v/o projections appear for exactly 4 of 52 layers and the remaining 48 carry Mamba-2 mixer tensors. Safetensors total 8100852736.

Notes

config.json (NemotronHForCausalLM, 52 layers, hybrid_override_pattern 'M-M-M-M*-...' with four starred positions) plus model.safetensors.index.json, in which q/k/v/o projections appear for exactly 4 of 52 layers and the remaining 48 carry Mamba-2 mixer tensors. Safetensors total 8100852736.

Cross-references

This record references no other records.