Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,15 +1,30 @@
|
|
1 |
# π Masked Word Predictor | CPU-only HF Space
|
2 |
|
3 |
import gradio as gr
|
4 |
-
from transformers import pipeline
|
5 |
|
6 |
# Load the fill-mask pipeline once at startup
|
7 |
fill_mask = pipeline("fill-mask", model="distilroberta-base", device=-1)
|
8 |
|
9 |
def predict_mask(sentence: str, top_k: int):
|
10 |
-
|
11 |
-
|
12 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
return [
|
14 |
{"sequence": p["sequence"], "score": round(p["score"], 3)}
|
15 |
for p in preds
|
@@ -18,20 +33,21 @@ def predict_mask(sentence: str, top_k: int):
|
|
18 |
with gr.Blocks(title="π Masked Word Predictor") as demo:
|
19 |
gr.Markdown(
|
20 |
"# π Masked Word Predictor\n"
|
21 |
-
"Enter a sentence with one `[MASK]` token and see the model
|
22 |
)
|
23 |
|
24 |
with gr.Row():
|
25 |
sentence = gr.Textbox(
|
26 |
lines=2,
|
27 |
-
placeholder="The
|
28 |
label="Input Sentence"
|
29 |
)
|
30 |
top_k = gr.Slider(
|
31 |
-
minimum=1, maximum=10,
|
32 |
label="Top K Predictions"
|
33 |
)
|
34 |
-
|
|
|
35 |
|
36 |
results = gr.Dataframe(
|
37 |
headers=["sequence", "score"],
|
|
|
1 |
# π Masked Word Predictor | CPU-only HF Space
|
2 |
|
3 |
import gradio as gr
|
4 |
+
from transformers import pipeline, PipelineException
|
5 |
|
6 |
# Load the fill-mask pipeline once at startup
|
7 |
fill_mask = pipeline("fill-mask", model="distilroberta-base", device=-1)
|
8 |
|
9 |
def predict_mask(sentence: str, top_k: int):
|
10 |
+
# Get the modelβs actual mask token (e.g. "<mask>")
|
11 |
+
mask = fill_mask.tokenizer.mask_token
|
12 |
+
|
13 |
+
# Allow users to type [MASK]; convert it under the hood
|
14 |
+
if "[MASK]" in sentence:
|
15 |
+
sentence = sentence.replace("[MASK]", mask)
|
16 |
+
|
17 |
+
# If no mask token present, show error
|
18 |
+
if mask not in sentence:
|
19 |
+
return [{"sequence": f"Error: please include `[MASK]` in your sentence.", "score": 0.0}]
|
20 |
+
|
21 |
+
# Call the pipeline and catch any unexpected exceptions
|
22 |
+
try:
|
23 |
+
preds = fill_mask(sentence, top_k=top_k)
|
24 |
+
except PipelineException as e:
|
25 |
+
return [{"sequence": f"Error: {str(e)}", "score": 0.0}]
|
26 |
+
|
27 |
+
# Format into list-of-dicts for Gradio Dataframe
|
28 |
return [
|
29 |
{"sequence": p["sequence"], "score": round(p["score"], 3)}
|
30 |
for p in preds
|
|
|
33 |
with gr.Blocks(title="π Masked Word Predictor") as demo:
|
34 |
gr.Markdown(
|
35 |
"# π Masked Word Predictor\n"
|
36 |
+
"Enter a sentence with one `[MASK]` token and see the top-K model predictions."
|
37 |
)
|
38 |
|
39 |
with gr.Row():
|
40 |
sentence = gr.Textbox(
|
41 |
lines=2,
|
42 |
+
placeholder="e.g. The salonβs new color treatment is [MASK].",
|
43 |
label="Input Sentence"
|
44 |
)
|
45 |
top_k = gr.Slider(
|
46 |
+
minimum=1, maximum=10, value=5, step=1,
|
47 |
label="Top K Predictions"
|
48 |
)
|
49 |
+
|
50 |
+
predict_btn = gr.Button("Predict π", variant="primary")
|
51 |
|
52 |
results = gr.Dataframe(
|
53 |
headers=["sequence", "score"],
|