Spaces:
Running
on
Zero
Running
on
Zero
Prgckwb
commited on
Commit
·
e613ea6
1
Parent(s):
2d945ae
change
Browse files
app.py
CHANGED
@@ -1,35 +1,38 @@
|
|
1 |
import os
|
|
|
2 |
from collections import Counter
|
3 |
|
4 |
import gradio as gr
|
5 |
import polars as pl
|
6 |
import spaces
|
7 |
import torch
|
8 |
-
import random
|
9 |
|
10 |
from metric import PerplexityCalculator
|
11 |
|
12 |
-
IS_DEBUG =
|
13 |
|
14 |
-
os.environ[
|
15 |
-
os.environ[
|
16 |
PAD_TOKEN_LABEL_ID = torch.nn.CrossEntropyLoss().ignore_index
|
17 |
-
DEVICE = torch.device(
|
18 |
|
19 |
-
df_sample_submission = pl.read_csv(
|
20 |
-
text_list = df_sample_submission.get_column(
|
21 |
text_counters = [Counter(text.split()) for text in text_list]
|
22 |
|
23 |
# Model Loading
|
24 |
if not IS_DEBUG:
|
25 |
-
scorer = PerplexityCalculator(
|
26 |
|
27 |
|
28 |
@spaces.GPU()
|
29 |
def inference(text: str, progress=gr.Progress(track_tqdm=True)):
|
|
|
|
|
30 |
if IS_DEBUG:
|
31 |
-
|
32 |
else:
|
|
|
33 |
score = scorer.get_perplexity(text)
|
34 |
|
35 |
input_counter = Counter(text.split())
|
@@ -37,27 +40,28 @@ def inference(text: str, progress=gr.Progress(track_tqdm=True)):
|
|
37 |
|
38 |
if any(is_match_list):
|
39 |
index = is_match_list.index(True)
|
40 |
-
index_text
|
41 |
return score, index_text
|
42 |
else:
|
43 |
-
index_text
|
44 |
gr.Warning(index_text)
|
45 |
return score, index_text
|
46 |
|
|
|
47 |
def random_inference(text: str, progress=gr.Progress(track_tqdm=True)):
|
48 |
-
if text ==
|
49 |
text = text_list[0]
|
50 |
|
51 |
words = text.split()
|
52 |
random.shuffle(words)
|
53 |
|
54 |
-
random_text =
|
55 |
|
56 |
score, index_text = inference(random_text)
|
57 |
return random_text, score, index_text
|
58 |
|
59 |
|
60 |
-
if __name__ ==
|
61 |
theme = gr.themes.Default(
|
62 |
primary_hue=gr.themes.colors.emerald,
|
63 |
secondary_hue=gr.themes.colors.green,
|
@@ -66,28 +70,41 @@ if __name__ == '__main__':
|
|
66 |
with gr.Blocks(theme=theme) as demo:
|
67 |
with gr.Column():
|
68 |
title = gr.Markdown(
|
69 |
-
"<h1 style='text-align: center; margin-bottom: 1rem'>Gemma-2-9b Perplexity Calculator</h1>"
|
|
|
70 |
|
71 |
with gr.Row():
|
72 |
with gr.Column():
|
73 |
-
input_text = gr.Textbox(label=
|
74 |
|
75 |
-
output_perplexity = gr.Number(label=
|
76 |
-
output_index = gr.Textbox(label=
|
77 |
|
78 |
with gr.Row():
|
79 |
-
clear_button = gr.ClearButton(
|
80 |
-
|
81 |
-
|
|
|
|
|
82 |
|
83 |
with gr.Column():
|
84 |
output_perplexity.render()
|
85 |
output_index.render()
|
86 |
|
87 |
-
sample_table = gr.Dataframe(
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
92 |
|
93 |
demo.queue().launch()
|
|
|
1 |
import os
|
2 |
+
import random
|
3 |
from collections import Counter
|
4 |
|
5 |
import gradio as gr
|
6 |
import polars as pl
|
7 |
import spaces
|
8 |
import torch
|
|
|
9 |
|
10 |
from metric import PerplexityCalculator
|
11 |
|
12 |
+
IS_DEBUG = False
|
13 |
|
14 |
+
os.environ["OMP_NUM_THREADS"] = "1"
|
15 |
+
os.environ["TOKENIZERS_PARALLELISM"] = "false"
|
16 |
PAD_TOKEN_LABEL_ID = torch.nn.CrossEntropyLoss().ignore_index
|
17 |
+
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
18 |
|
19 |
+
df_sample_submission = pl.read_csv("data/sample_submission.csv")
|
20 |
+
text_list = df_sample_submission.get_column("text").to_list()
|
21 |
text_counters = [Counter(text.split()) for text in text_list]
|
22 |
|
23 |
# Model Loading
|
24 |
if not IS_DEBUG:
|
25 |
+
scorer = PerplexityCalculator("google/gemma-2-9b")
|
26 |
|
27 |
|
28 |
@spaces.GPU()
|
29 |
def inference(text: str, progress=gr.Progress(track_tqdm=True)):
|
30 |
+
score = -1
|
31 |
+
|
32 |
if IS_DEBUG:
|
33 |
+
index_text = f"[DEBUG] "
|
34 |
else:
|
35 |
+
index_text = ""
|
36 |
score = scorer.get_perplexity(text)
|
37 |
|
38 |
input_counter = Counter(text.split())
|
|
|
40 |
|
41 |
if any(is_match_list):
|
42 |
index = is_match_list.index(True)
|
43 |
+
index_text += f"Task #{index}"
|
44 |
return score, index_text
|
45 |
else:
|
46 |
+
index_text += "No Match to All Tasks"
|
47 |
gr.Warning(index_text)
|
48 |
return score, index_text
|
49 |
|
50 |
+
|
51 |
def random_inference(text: str, progress=gr.Progress(track_tqdm=True)):
|
52 |
+
if text == "":
|
53 |
text = text_list[0]
|
54 |
|
55 |
words = text.split()
|
56 |
random.shuffle(words)
|
57 |
|
58 |
+
random_text = " ".join(words)
|
59 |
|
60 |
score, index_text = inference(random_text)
|
61 |
return random_text, score, index_text
|
62 |
|
63 |
|
64 |
+
if __name__ == "__main__":
|
65 |
theme = gr.themes.Default(
|
66 |
primary_hue=gr.themes.colors.emerald,
|
67 |
secondary_hue=gr.themes.colors.green,
|
|
|
70 |
with gr.Blocks(theme=theme) as demo:
|
71 |
with gr.Column():
|
72 |
title = gr.Markdown(
|
73 |
+
"<h1 style='text-align: center; margin-bottom: 1rem'>Gemma-2-9b Perplexity Calculator</h1>"
|
74 |
+
)
|
75 |
|
76 |
with gr.Row():
|
77 |
with gr.Column():
|
78 |
+
input_text = gr.Textbox(label="Text Input")
|
79 |
|
80 |
+
output_perplexity = gr.Number(label="Perplexity", render=False)
|
81 |
+
output_index = gr.Textbox(label="Index", render=False)
|
82 |
|
83 |
with gr.Row():
|
84 |
+
clear_button = gr.ClearButton(
|
85 |
+
[input_text, output_perplexity, output_index]
|
86 |
+
)
|
87 |
+
random_button = gr.Button("Randomize", variant="secondary")
|
88 |
+
submit_button = gr.Button("Run", variant="primary")
|
89 |
|
90 |
with gr.Column():
|
91 |
output_perplexity.render()
|
92 |
output_index.render()
|
93 |
|
94 |
+
sample_table = gr.Dataframe(
|
95 |
+
df_sample_submission, label="Sample Submission", type="polars"
|
96 |
+
)
|
97 |
+
|
98 |
+
submit_button.click(
|
99 |
+
inference, inputs=[input_text], outputs=[output_perplexity, output_index]
|
100 |
+
)
|
101 |
+
input_text.submit(
|
102 |
+
inference, inputs=[input_text], outputs=[output_perplexity, output_index]
|
103 |
+
)
|
104 |
+
random_button.click(
|
105 |
+
random_inference,
|
106 |
+
inputs=[input_text],
|
107 |
+
outputs=[input_text, output_perplexity, output_index],
|
108 |
+
)
|
109 |
|
110 |
demo.queue().launch()
|