Spaces:
Running
on
Zero
Running
on
Zero
Commit
•
9e7216d
1
Parent(s):
373e797
fetch data for toxicity if it doesn't exist yet
Browse files
app.py
CHANGED
@@ -10,7 +10,7 @@ import gradio as gr
|
|
10 |
import pandas as pd
|
11 |
import polars as pl
|
12 |
import matplotlib.pyplot as plt
|
13 |
-
import spaces
|
14 |
from gradio_huggingfacehub_search import HuggingfaceHubSearch
|
15 |
from huggingface_hub import PyTorchModelHubMixin
|
16 |
import torch
|
@@ -50,7 +50,7 @@ model = QualityModel.from_pretrained("nvidia/quality-classifier-deberta").to(dev
|
|
50 |
model.eval()
|
51 |
|
52 |
|
53 |
-
@spaces.GPU
|
54 |
def predict(texts: list[str]):
|
55 |
inputs = tokenizer(
|
56 |
texts, return_tensors="pt", padding="longest", truncation=True
|
@@ -81,7 +81,11 @@ def plot_and_df(texts, preds):
|
|
81 |
)
|
82 |
|
83 |
|
84 |
-
|
|
|
|
|
|
|
|
|
85 |
def run_quality_check(dataset, config, split, column, batch_size, num_examples):
|
86 |
logging.info(f"Fetching data for {dataset=} {config=} {split=} {column=}")
|
87 |
try:
|
@@ -97,9 +101,8 @@ def run_quality_check(dataset, config, split, column, batch_size, num_examples):
|
|
97 |
return
|
98 |
logging.info("Data fetched.")
|
99 |
|
100 |
-
|
101 |
-
|
102 |
-
# batch_size = 100
|
103 |
predictions, texts_processed = [], []
|
104 |
num_examples = min(len(texts), num_examples)
|
105 |
for i in range(0, num_examples, batch_size):
|
@@ -118,7 +121,7 @@ def run_quality_check(dataset, config, split, column, batch_size, num_examples):
|
|
118 |
# plt.xlabel('Proportion of non-ASCII characters')
|
119 |
# plt.ylabel('Number of texts')
|
120 |
|
121 |
-
yield {"finished": 1.}, *plot_and_df(texts_processed, predictions),
|
122 |
|
123 |
|
124 |
PERSPECTIVE_API_KEY = os.environ.get("PERSPECTIVE_API_KEY")
|
@@ -141,13 +144,31 @@ def plot_toxicity(scores):
|
|
141 |
|
142 |
return fig
|
143 |
|
144 |
-
def call_perspective_api(texts_df, column_name, full_check=False):
|
145 |
headers = {
|
146 |
"content-type": "application/json",
|
147 |
}
|
148 |
req_att_scores = {attr: [] for attr in REQUESTED_ATTRIBUTES}
|
149 |
|
150 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
151 |
|
152 |
n_samples = len(texts)
|
153 |
for i, text in tqdm(enumerate(texts), desc="scanning with perspective"):
|
@@ -165,8 +186,6 @@ def call_perspective_api(texts_df, column_name, full_check=False):
|
|
165 |
|
166 |
if req_response.ok:
|
167 |
response = req_response.json()
|
168 |
-
# logger.info("Perspective API response is:")
|
169 |
-
# logger.info(response)
|
170 |
if ATT_SCORE in response:
|
171 |
for req_att in REQUESTED_ATTRIBUTES:
|
172 |
if req_att in response[ATT_SCORE]:
|
@@ -175,15 +194,12 @@ def call_perspective_api(texts_df, column_name, full_check=False):
|
|
175 |
else:
|
176 |
req_att_scores[req_att].append(0)
|
177 |
else:
|
178 |
-
# logger.error(
|
179 |
-
# "Unexpected response format from Perspective API."
|
180 |
-
# )
|
181 |
raise ValueError(req_response)
|
182 |
else:
|
183 |
try:
|
184 |
req_response.raise_for_status()
|
185 |
except Exception as e:
|
186 |
-
|
187 |
return req_att_scores
|
188 |
if i % 10 == 0:
|
189 |
plot_toxicity(req_att_scores)
|
@@ -295,11 +311,9 @@ with gr.Blocks() as demo:
|
|
295 |
def show_input_from_split_dropdown(dataset: str, subset: str, split: str) -> dict:
|
296 |
return _resolve_dataset_selection(dataset, default_subset=subset, default_split=split)
|
297 |
|
298 |
-
# text_column = gr.Textbox(placeholder="text", label="Text colum name to check (data must be non-nested, raw texts!)")
|
299 |
-
|
300 |
gr.Markdown("## Run nvidia quality classifier")
|
301 |
batch_size = gr.Slider(0, 64, 32, step=4, label="Inference batch size (set this to smaller value if this space crashes.)")
|
302 |
-
num_examples = gr.
|
303 |
gr_check_btn = gr.Button("Check Dataset")
|
304 |
progress_bar = gr.Label(show_label=False)
|
305 |
plot = gr.BarPlot()
|
@@ -329,7 +343,7 @@ with gr.Blocks() as demo:
|
|
329 |
# gr_ascii_btn.click(non_ascii_check, inputs=[texts_df, text_column], outputs=[non_ascii_hist])
|
330 |
|
331 |
gr.Markdown("## Explore toxicity")
|
332 |
-
checkbox = gr.Checkbox(value=False, label="Run on full first parquet data (better not)")
|
333 |
gr_toxicity_btn = gr.Button("Run perpspective API to check toxicity of random samples.")
|
334 |
toxicity_progress_bar = gr.Label(show_label=False)
|
335 |
toxicity_hist = gr.Plot()
|
@@ -337,7 +351,7 @@ with gr.Blocks() as demo:
|
|
337 |
toxicity_df = gr.DataFrame()
|
338 |
gr_toxicity_btn.click(
|
339 |
call_perspective_api,
|
340 |
-
inputs=[texts_df, text_column_dropdown, checkbox],
|
341 |
outputs=[toxicity_progress_bar, toxicity_hist, toxicity_df]
|
342 |
)
|
343 |
|
|
|
10 |
import pandas as pd
|
11 |
import polars as pl
|
12 |
import matplotlib.pyplot as plt
|
13 |
+
# import spaces
|
14 |
from gradio_huggingfacehub_search import HuggingfaceHubSearch
|
15 |
from huggingface_hub import PyTorchModelHubMixin
|
16 |
import torch
|
|
|
50 |
model.eval()
|
51 |
|
52 |
|
53 |
+
# @spaces.GPU
|
54 |
def predict(texts: list[str]):
|
55 |
inputs = tokenizer(
|
56 |
texts, return_tensors="pt", padding="longest", truncation=True
|
|
|
81 |
)
|
82 |
|
83 |
|
84 |
+
# def download_data(dataset, config, split, column):
|
85 |
+
#
|
86 |
+
|
87 |
+
|
88 |
+
# @spaces.GPU
|
89 |
def run_quality_check(dataset, config, split, column, batch_size, num_examples):
|
90 |
logging.info(f"Fetching data for {dataset=} {config=} {split=} {column=}")
|
91 |
try:
|
|
|
101 |
return
|
102 |
logging.info("Data fetched.")
|
103 |
|
104 |
+
data_sample = data.sample(num_examples, seed=16) if data.shape[0] > num_examples else data
|
105 |
+
texts = [text[:10000] for text in data_sample[column].to_list()]
|
|
|
106 |
predictions, texts_processed = [], []
|
107 |
num_examples = min(len(texts), num_examples)
|
108 |
for i in range(0, num_examples, batch_size):
|
|
|
121 |
# plt.xlabel('Proportion of non-ASCII characters')
|
122 |
# plt.ylabel('Number of texts')
|
123 |
|
124 |
+
yield {"finished": 1.}, *plot_and_df(texts_processed, predictions), data_sample
|
125 |
|
126 |
|
127 |
PERSPECTIVE_API_KEY = os.environ.get("PERSPECTIVE_API_KEY")
|
|
|
144 |
|
145 |
return fig
|
146 |
|
147 |
+
def call_perspective_api(texts_df, column_name, dataset, config, split):#, full_check=False):
|
148 |
headers = {
|
149 |
"content-type": "application/json",
|
150 |
}
|
151 |
req_att_scores = {attr: [] for attr in REQUESTED_ATTRIBUTES}
|
152 |
|
153 |
+
# fetch data if it doesn't exist yet
|
154 |
+
if texts_df.values.tolist() == [['', '', '']]:
|
155 |
+
logging.info(f"Fetching data for {dataset=} {config=} {split=} {column_name=}")
|
156 |
+
try:
|
157 |
+
texts_df = pl.read_parquet(f"hf://datasets/{dataset}@~parquet/{config}/{split}/0000.parquet", columns=[column_name])
|
158 |
+
except pl.exceptions.ComputeError:
|
159 |
+
try:
|
160 |
+
texts_df = pl.read_parquet(f"hf://datasets/{dataset}@~parquet/{config}/partial-{split}/0000.parquet", columns=[column_name])
|
161 |
+
except pl.exceptions.ComputeError:
|
162 |
+
try:
|
163 |
+
texts_df = pl.read_parquet(f"hf://datasets/{dataset}@~parquet/{config}/{split}-part0/0000.parquet", columns=[column_name])
|
164 |
+
except Exception as error:
|
165 |
+
yield f"❌ {error}", plt.gcf(), pd.DataFrame(),
|
166 |
+
return
|
167 |
+
logging.info("Data fetched.")
|
168 |
+
texts_df = texts_df.to_pandas()
|
169 |
+
|
170 |
+
# texts = texts_df.sample(100, seed=16)[column_name].values if not full_check else texts_df[column_name].values
|
171 |
+
texts = texts_df.sample(100, random_state=16)[column_name].values if texts_df.shape[0] > 100 else texts_df[column_name].values
|
172 |
|
173 |
n_samples = len(texts)
|
174 |
for i, text in tqdm(enumerate(texts), desc="scanning with perspective"):
|
|
|
186 |
|
187 |
if req_response.ok:
|
188 |
response = req_response.json()
|
|
|
|
|
189 |
if ATT_SCORE in response:
|
190 |
for req_att in REQUESTED_ATTRIBUTES:
|
191 |
if req_att in response[ATT_SCORE]:
|
|
|
194 |
else:
|
195 |
req_att_scores[req_att].append(0)
|
196 |
else:
|
|
|
|
|
|
|
197 |
raise ValueError(req_response)
|
198 |
else:
|
199 |
try:
|
200 |
req_response.raise_for_status()
|
201 |
except Exception as e:
|
202 |
+
logging.info(e)
|
203 |
return req_att_scores
|
204 |
if i % 10 == 0:
|
205 |
plot_toxicity(req_att_scores)
|
|
|
311 |
def show_input_from_split_dropdown(dataset: str, subset: str, split: str) -> dict:
|
312 |
return _resolve_dataset_selection(dataset, default_subset=subset, default_split=split)
|
313 |
|
|
|
|
|
314 |
gr.Markdown("## Run nvidia quality classifier")
|
315 |
batch_size = gr.Slider(0, 64, 32, step=4, label="Inference batch size (set this to smaller value if this space crashes.)")
|
316 |
+
num_examples = gr.Slider(0, 1000, 500, step=10, label="Number of random examples to check")
|
317 |
gr_check_btn = gr.Button("Check Dataset")
|
318 |
progress_bar = gr.Label(show_label=False)
|
319 |
plot = gr.BarPlot()
|
|
|
343 |
# gr_ascii_btn.click(non_ascii_check, inputs=[texts_df, text_column], outputs=[non_ascii_hist])
|
344 |
|
345 |
gr.Markdown("## Explore toxicity")
|
346 |
+
# checkbox = gr.Checkbox(value=False, label="Run on full first parquet data (better not)")
|
347 |
gr_toxicity_btn = gr.Button("Run perpspective API to check toxicity of random samples.")
|
348 |
toxicity_progress_bar = gr.Label(show_label=False)
|
349 |
toxicity_hist = gr.Plot()
|
|
|
351 |
toxicity_df = gr.DataFrame()
|
352 |
gr_toxicity_btn.click(
|
353 |
call_perspective_api,
|
354 |
+
inputs=[texts_df, text_column_dropdown, dataset_name, subset_dropdown, split_dropdown],#, checkbox],
|
355 |
outputs=[toxicity_progress_bar, toxicity_hist, toxicity_df]
|
356 |
)
|
357 |
|