File size: 600 Bytes
9f8613e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
---
license: apache-2.0
base_model: fal/AuraFlow-v0.3
base_model_relation: quantized
---


FP8 quantized version of [AuraFlow v0.3](fal/AuraFlow-v0.3)

## Quantization

```py
import torch
from huggingface_hub import cached_download
from safetensors.torch import load_file, save_file

ckpt_path = cached_download(
    "https://huggingface.co/fal/AuraFlow-v0.3/resolve/main/aura_flow_0.3.safetensors",
)

state_dict = load_file(ckpt_path)

for key, value in state_dict.items():
    state_dict[key] = value.to(torch.float8_e4m3fn)

save_file(state_dict, "./aura_flow_0.3.float8_e4m3fn.safetensors")
```