Spaces:
Sleeping
Sleeping
Shing Yee
commited on
Update application
Browse files
app.py
CHANGED
@@ -12,21 +12,25 @@ from utils import (
|
|
12 |
cross_encoder_predict_relevance
|
13 |
)
|
14 |
|
15 |
-
def predict(system_prompt, user_prompt
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
elif selected_model == "cross-encoder/ms-marco-MiniLM-L-6-v2":
|
21 |
-
predicted_label, probabilities = cross_encoder_predict_relevance(system_prompt, user_prompt, ms_model, ms_tokenizer, device)
|
22 |
-
|
23 |
-
probability_off_topic = probabilities[0][1] * 100
|
24 |
-
label = "Off-topic" if predicted_label==1 else "On-topic"
|
25 |
result = f"""
|
26 |
-
**Prediction Summary
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
|
28 |
-
|
29 |
-
- **
|
|
|
30 |
"""
|
31 |
|
32 |
return result
|
@@ -36,15 +40,8 @@ with gr.Blocks(theme=gr.themes.Soft(), fill_height=True) as app:
|
|
36 |
gr.Markdown("# Off-Topic Classification using Fine-tuned Embeddings and Cross-Encoder Models")
|
37 |
|
38 |
with gr.Row():
|
39 |
-
system_prompt = gr.
|
40 |
-
user_prompt = gr.
|
41 |
-
|
42 |
-
with gr.Row():
|
43 |
-
selected_model = gr.Dropdown(
|
44 |
-
["jinaai/jina-embeddings-v2-small-en",
|
45 |
-
"cross-encoder/stsb-roberta-base",
|
46 |
-
"cross-encoder/ms-marco-MiniLM-L-6-v2"],
|
47 |
-
label="Select a model")
|
48 |
|
49 |
# Button to run the prediction
|
50 |
get_classfication = gr.Button("Check Content")
|
@@ -53,7 +50,7 @@ with gr.Blocks(theme=gr.themes.Soft(), fill_height=True) as app:
|
|
53 |
|
54 |
get_classfication.click(
|
55 |
fn=predict,
|
56 |
-
inputs=[system_prompt, user_prompt
|
57 |
outputs=output_result
|
58 |
)
|
59 |
|
|
|
12 |
cross_encoder_predict_relevance
|
13 |
)
|
14 |
|
15 |
+
def predict(system_prompt, user_prompt):
|
16 |
+
predicted_label_jina, probabilities_jina = embeddings_predict_relevance(system_prompt, user_prompt, jina_model, jina_tokenizer, device)
|
17 |
+
predicted_label_stsb, probabilities_stsb = cross_encoder_predict_relevance(system_prompt, user_prompt, stsb_model, stsb_tokenizer, device)
|
18 |
+
predicted_label_ms, probabilities_ms = cross_encoder_predict_relevance(system_prompt, user_prompt, ms_model, ms_tokenizer, device)
|
19 |
+
|
|
|
|
|
|
|
|
|
|
|
20 |
result = f"""
|
21 |
+
**Prediction Summary**
|
22 |
+
|
23 |
+
**1. Model: jinaai/jina-embeddings-v2-small-en**
|
24 |
+
- **Prediction**: {"π₯ Off-topic" if predicted_label_jina==1 else "π© On-topic"}
|
25 |
+
- **Probability of being off-topic**: {probabilities_jina[0][1]:.2%}
|
26 |
+
|
27 |
+
**2. Model: cross-encoder/stsb-roberta-base**
|
28 |
+
- **Prediction**: {"π₯ Off-topic" if predicted_label_stsb==1 else "π© On-topic"}
|
29 |
+
- **Probability of being off-topic**: {probabilities_stsb[0][1]:.2%}
|
30 |
|
31 |
+
**3. Model: cross-encoder/ms-marco-MiniLM-L-6-v2**
|
32 |
+
- **Prediction**: {"π₯ Off-topic" if predicted_label_ms==1 else "π© On-topic"}
|
33 |
+
- **Probability of being off-topic**: {probabilities_ms[0][1]:.2%}
|
34 |
"""
|
35 |
|
36 |
return result
|
|
|
40 |
gr.Markdown("# Off-Topic Classification using Fine-tuned Embeddings and Cross-Encoder Models")
|
41 |
|
42 |
with gr.Row():
|
43 |
+
system_prompt = gr.TextArea(label="System Prompt", lines=5)
|
44 |
+
user_prompt = gr.TextArea(label="User Prompt", lines=5)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
45 |
|
46 |
# Button to run the prediction
|
47 |
get_classfication = gr.Button("Check Content")
|
|
|
50 |
|
51 |
get_classfication.click(
|
52 |
fn=predict,
|
53 |
+
inputs=[system_prompt, user_prompt],
|
54 |
outputs=output_result
|
55 |
)
|
56 |
|