p1atdev/Neta-Lumina-bnb-nf4

Neta-Lumina bitsandbytes NF4 quantization

File	Size
neta-lumina-v1.0-AIO.bnb_nf4.safetensors	3.75 GB

Quantization

• Quantization type: bitsandbytes NF4

• Denoiser (model.diffusion_model.)

  • Traget keys:
    • ".attention."
    • ".feed_forward."
    • ".adaLN_modulation."
  • Exclude keys:
    • ".final_layer."

• Text encoder (text_encoders.gemma2_2b.transformer.)

  • Target keys:
    • ".self_attn."
    • ".mlp."
  • Exclude: None

Samples

Default (BF16)	Text encoder NF4	Denoiser NF4	Text encoder & Denoiser NF4 (This repo)