C-RADIOv4-H-3.5-Super

This repository packages the C-RADIOv4-H Tower, final LayerNorm, and Projector from Nemotron 3.5 Super VL.

Contents

File	Tensors	What it holds
model.safetensors	391	RADIO Tower from Super vision_model.*, including summary_idxs
final_norm.safetensors	2	vision_projector.vision_final_layernorm
projector.safetensors	3	Super mlp1 Projector
config.json, inventory.json		Source and code pins, widths, and the shape, dtype and SHA256 of every tensor
vision.py		Loader used by examples/inference.py

Architecture

Component	Details
Tower	C-RADIOv4-H, 32 layers, 1280 hidden, patch size 16, ten prefix tokens
Final LayerNorm	LayerNorm(1280), eps 1e-6, applied before the pixel shuffle
Token compression	InternVL v2 2x2 pixel shuffle, scale 0.5, no learned parameters
Projector (mlp1)	RMSNorm(5120), Linear(5120, 20480) no bias, SquaredReLU, Linear(20480, 4096) no bias

At 448 the bench reports 784 Tower tokens of width 1280, 196 merged tokens of width 5120 (after the final LayerNorm), and 196 projected tokens of width 4096.

Usage

See examples/inference.py for image feature extraction.

Validation

The release parity test compares every exported tensor with the pinned parent using torch.equal, then checks all stages against the parent's own RadioModel and VisionProjector code.

Reproduction

The export script reads vision_model.*, mlp1.*, and vision_projector.vision_final_layernorm.* from the pinned parent and writes the original BF16 tensors.

Credits

NVIDIA released Nemotron 3.5 Super VL and C-RADIOv4-H.

License

OpenMDW-1.1, the license from the source model card.