Update README.md
Browse files
README.md
CHANGED
|
@@ -1,5 +1,26 @@
|
|
| 1 |
-
---
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: diffusers
|
| 3 |
+
license: other
|
| 4 |
+
license_name: flux-1-dev-non-commercial-license
|
| 5 |
+
license_link: https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md
|
| 6 |
+
---
|
| 7 |
+
|
| 8 |
+
# How to cast fp8
|
| 9 |
+
1. Install quanto-optinum.
|
| 10 |
+
```bash
|
| 11 |
+
pip install optimum-quanto
|
| 12 |
+
```
|
| 13 |
+
2. Run the script:
|
| 14 |
+
```python
|
| 15 |
+
import torch
|
| 16 |
+
from safetensors.torch import save_file
|
| 17 |
+
|
| 18 |
+
from diffusers import FluxTransformer2DModel, FluxPipeline
|
| 19 |
+
from optimum.quanto import freeze, qfloat8, quantize
|
| 20 |
+
|
| 21 |
+
transformer = FluxTransformer2DModel.from_single_file("modern-anime.safetensors", torch_dtype=torch.bfloat16)
|
| 22 |
+
quantize(transformer, weights=qfloat8)
|
| 23 |
+
freeze(transformer)
|
| 24 |
+
|
| 25 |
+
transformer.save_pretrained("transformer")
|
| 26 |
+
```
|