File size: 975 Bytes
9a70cb3
 
 
923ab5c
9a70cb3
923ab5c
 
9a70cb3
77bf2f9
183d759
 
 
 
9a70cb3
 
 
618804a
 
 
9a70cb3
 
 
a325776
9a70cb3
f790223
9a70cb3
61f602a
9a70cb3
 
 
 
ca1c3d5
2838c66
f790223
 
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
29
30
31
32
33
34
35
import gradio as gr
from fastai.vision.all import *
from huggingface_hub import from_pretrained_fastai
import torch, os

os.environ.setdefault("OMP_NUM_THREADS", "1")
torch.set_num_threads(1)

learn = from_pretrained_fastai("Pablogps/castle-classifier-25")
try:
    learn.to_fp32()
except:
    pass
labels = learn.dls.vocab

def predict(img):
    img = PILImage.create(img)          # same flow as before
    pred, pred_idx, probs = learn.predict(img)
    return {labels[i]: float(probs[i]) for i in range(len(labels))}

title = "Bad castle predictor"
description = "A bad model that tries to identify the type of castle."
examples = ['spanish', 'french', 'japanese']

demo = gr.Interface(
    fn=predict,
    inputs=gr.Image(type="pil"),
    outputs=gr.Label(num_top_classes=3),
    title=title,
    description=description,
    examples=examples,
    cache_examples=False,       # <-- don’t pre-run at startup
)

demo.queue(max_size=8).launch(show_error=True, debug=True)