---
license: openrail++
base_model:
- cagliostrolab/animagine-xl-4.0
base_model_relation: quantized
tags:
- bitsandbytes
---

## Quantization settings

- `vae` (`first_stage_model`): `torch.float16`. No quantization.
- `text_encoder`, `text_encoder_2` (`conditioner.embedders`):
  - NF4 with [bitsandbytes](https://github.com/bitsandbytes-foundation/bitsandbytes)
  - Target layers:`["self_attn", ".mlp."]`
- `diffusion_model`:
  - Int8 with [bitsandbytes](https://github.com/bitsandbytes-foundation/bitsandbytes)
  - Target layers: `["attn1", "attn2", ".ff."]`