Spaces:
Sleeping
Sleeping
for team
Browse files- .gitignore +1 -9
- README.md +6 -6
- app.py +441 -0
- note.txt +1 -1
- requirements.txt +1 -2
.gitignore
CHANGED
@@ -161,12 +161,4 @@ cython_debug/
|
|
161 |
# option (not recommended) you can uncomment the following to ignore the entire idea folder.
|
162 |
#.idea/
|
163 |
|
164 |
-
user_annotation/*
|
165 |
-
|
166 |
-
run_2.py
|
167 |
-
run_3.py
|
168 |
-
run_4.py
|
169 |
-
backup.py
|
170 |
-
idea.txt
|
171 |
-
|
172 |
-
dataclass.py
|
|
|
161 |
# option (not recommended) you can uncomment the following to ignore the entire idea folder.
|
162 |
#.idea/
|
163 |
|
164 |
+
user_annotation/*
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
README.md
CHANGED
@@ -1,12 +1,12 @@
|
|
1 |
---
|
2 |
-
title:
|
3 |
-
emoji:
|
4 |
-
colorFrom:
|
5 |
-
colorTo:
|
6 |
sdk: gradio
|
7 |
-
app_file:
|
8 |
pinned: false
|
9 |
-
license:
|
10 |
|
11 |
hf_oauth: true
|
12 |
# optional, default duration is 8 hours/480 minutes. Max duration is 30 days/43200 minutes.
|
|
|
1 |
---
|
2 |
+
title: Tanuki Annotation Phase2
|
3 |
+
emoji: 📊
|
4 |
+
colorFrom: red
|
5 |
+
colorTo: red
|
6 |
sdk: gradio
|
7 |
+
app_file: app.py
|
8 |
pinned: false
|
9 |
+
license: unknown
|
10 |
|
11 |
hf_oauth: true
|
12 |
# optional, default duration is 8 hours/480 minutes. Max duration is 30 days/43200 minutes.
|
app.py
ADDED
@@ -0,0 +1,441 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
|
3 |
+
import json
|
4 |
+
import datetime
|
5 |
+
from pathlib import Path
|
6 |
+
import uuid
|
7 |
+
from typing import Tuple
|
8 |
+
|
9 |
+
import pandas as pd
|
10 |
+
|
11 |
+
import gradio as gr
|
12 |
+
from datasets import load_dataset
|
13 |
+
from huggingface_hub import CommitScheduler
|
14 |
+
from huggingface_hub import HfFolder
|
15 |
+
|
16 |
+
|
17 |
+
# HF_Spaceでプライベート関連にアクセスするための環境変数
|
18 |
+
# SecretKey をSpaceのSettingsに設定
|
19 |
+
HF_TOKEN = os.getenv("HF_TOKEN")
|
20 |
+
if HF_TOKEN:
|
21 |
+
HfFolder.save_token(HF_TOKEN)
|
22 |
+
else:
|
23 |
+
print("Warning: HF_TOKEN not found. Please set it in your Space secrets.")
|
24 |
+
|
25 |
+
# HFデータセット アップロード先
|
26 |
+
# (切替てテストする用に配列)
|
27 |
+
output_dataset = [
|
28 |
+
"team-hatakeyama-phase2/annotation_tanuki_phase2",
|
29 |
+
"kevineen/Tanuki-Phase2-annotation-dataset", # test
|
30 |
+
]
|
31 |
+
|
32 |
+
# アノテーション対象データセット
|
33 |
+
annotation_dataset_list = [
|
34 |
+
"hatakeyama-llm-team/AutoGeneratedJapaneseQA",
|
35 |
+
"hatakeyama-llm-team/AutoGeneratedJapaneseQA-other",
|
36 |
+
"kanhatakeyama/ChatbotArenaJaMixtral8x22b",
|
37 |
+
"kanhatakeyama/OrcaJaMixtral8x22b",
|
38 |
+
"kanhatakeyama/LogicalDatasetsByMixtral8x22b",
|
39 |
+
|
40 |
+
# データ形式未対応(対応予定
|
41 |
+
# "hatakeyama-llm-team/WikiBookJa",
|
42 |
+
# "kanhatakeyama/AutoWikiQA",
|
43 |
+
# "susumuota/SyntheticTextWikiTranslate-askllm-v1", # Ask-LLM 翻訳
|
44 |
+
]
|
45 |
+
|
46 |
+
multi_turn_annotation_dataset_list = [
|
47 |
+
# マルチターン 未対応
|
48 |
+
"kanhatakeyama/AutoMultiTurnByMixtral8x22b",
|
49 |
+
]
|
50 |
+
|
51 |
+
# Session State : (ブラウザセッション単位の変数管理) ===========================
|
52 |
+
|
53 |
+
# UIのEnable/Disable用State
|
54 |
+
is_selected_dataset = gr.State(False)
|
55 |
+
is_loaded_dataset = gr.State(False)
|
56 |
+
|
57 |
+
# 選択中のデータセットリスト
|
58 |
+
dropdown_dataset_list = gr.State(value=annotation_dataset_list)
|
59 |
+
# 現在の対象データセット 初期値は"hatakeyama-llm-team/AutoGeneratedJapaneseQA",
|
60 |
+
select_dropdown_dataset = gr.State(dropdown_dataset_list.value[0])
|
61 |
+
select_dataset = gr.State(None) # 現在のデータセット
|
62 |
+
select_dataset_total_len = gr.State(0) # 現在のデータセットの長さ
|
63 |
+
select_idx = gr.State(0) # 現在のインデックス (ランダムモードにするなら不要?
|
64 |
+
random_mode = gr.State(False)
|
65 |
+
|
66 |
+
# 回答者がアノテーションしたデータセット
|
67 |
+
annotated_dataset = gr.State(
|
68 |
+
pd.DataFrame({
|
69 |
+
'dataset_name': [],
|
70 |
+
'dataset_id': [],
|
71 |
+
'who': [],
|
72 |
+
'good': [],
|
73 |
+
'bad': [],
|
74 |
+
'score': [],
|
75 |
+
'is_proofreading_1': [],
|
76 |
+
'answer_text_1': [],
|
77 |
+
'is_proofreading_2': [], # マルチターン用
|
78 |
+
'answer_text_2': [], # マルチターン用
|
79 |
+
})
|
80 |
+
)
|
81 |
+
|
82 |
+
initial_answer_text_1 = gr.State("") # 回答1を整形したかチェック用
|
83 |
+
initial_answer_text_2 = gr.State("") # 回答2を整形したかチェック用
|
84 |
+
|
85 |
+
is_dataset_loaded = gr.State(False)
|
86 |
+
|
87 |
+
you_dataset_id = gr.State(0) # 回答者がアノテーションしているデータのID
|
88 |
+
dataset_name = gr.State("") # 編集に使用したデータセット名
|
89 |
+
dataset_id = gr.State(0) # 加工元データセットのindex
|
90 |
+
who = gr.State("") # アノテーション者名
|
91 |
+
good = gr.State(False) # 良
|
92 |
+
bad = gr.State(False) # 悪
|
93 |
+
score = gr.State(3) # スコア 初期値は3
|
94 |
+
is_proofreading_1 = gr.State(False) # 回答1を整形したか_1
|
95 |
+
answer_text_1 = gr.State("") # answer_1 回答
|
96 |
+
is_proofreading_2 = gr.State(False) # 回答2を整形したか_2
|
97 |
+
answer_text_2 = gr.State("") # answer_2 回答
|
98 |
+
|
99 |
+
# 未整理
|
100 |
+
# データ読み込み ========================================
|
101 |
+
|
102 |
+
def dataset_load_fn() -> Tuple[
|
103 |
+
str,
|
104 |
+
str,
|
105 |
+
str,
|
106 |
+
str,
|
107 |
+
gr.update,
|
108 |
+
gr.update,
|
109 |
+
gr.update,
|
110 |
+
gr.update,
|
111 |
+
gr.update,
|
112 |
+
gr.update,
|
113 |
+
gr.update]:
|
114 |
+
|
115 |
+
is_dataset_loaded.value = False # ロード状態
|
116 |
+
|
117 |
+
select_dataset.value = load_dataset(
|
118 |
+
select_dropdown_dataset.value
|
119 |
+
)
|
120 |
+
|
121 |
+
# DatasetオブジェクトをPandas DataFrameに変換
|
122 |
+
df = select_dataset.value["train"].to_pandas()
|
123 |
+
|
124 |
+
# index列を追加し、シャッフル
|
125 |
+
df = df.reset_index(drop=False) # 元のindexを保持
|
126 |
+
df = df.sample(frac=1).reset_index(drop=True) # シャッフル
|
127 |
+
select_dataset.value["train"] = df # シャッフルされたDataFrameを格納
|
128 |
+
|
129 |
+
select_idx.value = 0 # index初期化
|
130 |
+
select_dataset_total_len.value = len(df) # 長さを取得
|
131 |
+
is_dataset_loaded.value = True # ロード完了
|
132 |
+
|
133 |
+
# データロード時に初期値を設定
|
134 |
+
initial_answer_text_1.value = df.iloc[select_idx.value]["answer"]
|
135 |
+
initial_answer_text_2.value = df.iloc[select_idx.value]["answer"]
|
136 |
+
|
137 |
+
return df.iloc[select_idx.value]["question"], \
|
138 |
+
df.iloc[select_idx.value]["answer"], \
|
139 |
+
df.iloc[select_idx.value]["question"], \
|
140 |
+
df.iloc[select_idx.value]["answer"], \
|
141 |
+
gr.update(interactive=True), \
|
142 |
+
gr.update(interactive=True), \
|
143 |
+
gr.update(interactive=True), \
|
144 |
+
gr.update(interactive=True), \
|
145 |
+
gr.update(interactive=True), \
|
146 |
+
gr.update(interactive=True), \
|
147 |
+
gr.update(interactive=True)
|
148 |
+
|
149 |
+
|
150 |
+
# データの保存処理 ========================================
|
151 |
+
|
152 |
+
# Spaceの場合の保存先はCommitSchedulerのpath_in_repoフォルダ
|
153 |
+
# (ローカル開発の場合./user_annotationフォルダにjsonファイルが作成される)
|
154 |
+
annotation_file = Path("user_annotation/") / f"data_{uuid.uuid4()}.json"
|
155 |
+
annotated_folder = annotation_file.parent
|
156 |
+
|
157 |
+
scheduler = CommitScheduler(
|
158 |
+
repo_id=output_dataset[0],
|
159 |
+
repo_type="dataset",
|
160 |
+
folder_path=annotated_folder,
|
161 |
+
path_in_repo="data", # Spaceの場合の保存先フォルダー
|
162 |
+
private=True,
|
163 |
+
every=5, # 5分毎にアップロード HuggingFAce_Documentの最低推奨値
|
164 |
+
)
|
165 |
+
|
166 |
+
# CommitScheduler (HFへのデータアップロード
|
167 |
+
def save_annotation(
|
168 |
+
dataset_name: str,
|
169 |
+
dataset_id: int,
|
170 |
+
who: str,
|
171 |
+
good: bool,
|
172 |
+
bad: bool,
|
173 |
+
score: int,
|
174 |
+
is_proofreading_1: bool,
|
175 |
+
answer_text_1: str,
|
176 |
+
is_proofreading_2: bool,
|
177 |
+
answer_text_2: str) -> None:
|
178 |
+
|
179 |
+
annotated_dataset.value = pd.concat([
|
180 |
+
annotated_dataset.value,
|
181 |
+
pd.DataFrame({
|
182 |
+
'dataset_name': [dataset_name],
|
183 |
+
'dataset_id': [dataset_id],
|
184 |
+
'who': [who],
|
185 |
+
'good': [good],
|
186 |
+
'bad': [bad],
|
187 |
+
'score': [score],
|
188 |
+
'is_proofreading_1': [is_proofreading_1],
|
189 |
+
"answer_text_1": [answer_text_1],
|
190 |
+
'is_proofreading_2': [is_proofreading_2],
|
191 |
+
'answer_text_2': [answer_text_2]
|
192 |
+
})], ignore_index=True).reset_index(drop=True)
|
193 |
+
|
194 |
+
# 書き込み
|
195 |
+
with scheduler.lock:
|
196 |
+
with annotation_file.open("a", encoding='utf-8') as f:
|
197 |
+
data_to_write = {
|
198 |
+
# "id": , CommitSchedulerだと取得して末尾idを付与することが無理?
|
199 |
+
"datetime": str(datetime.datetime.now().isoformat()),
|
200 |
+
"dataset_name": dataset_name,
|
201 |
+
"dataset_id": int(dataset_id),
|
202 |
+
"who": who,
|
203 |
+
"good": good,
|
204 |
+
"bad": bad,
|
205 |
+
"score": score,
|
206 |
+
"is_proofreading_1": is_proofreading_1,
|
207 |
+
"answer_text_1": answer_text_1,
|
208 |
+
"is_proofreading_2": is_proofreading_2,
|
209 |
+
"answer_text_2": answer_text_2,
|
210 |
+
}
|
211 |
+
f.write(json.dumps(data_to_write, ensure_ascii=False))
|
212 |
+
f.write("\n")
|
213 |
+
|
214 |
+
# アノテーションの追加処理 ========================================
|
215 |
+
|
216 |
+
|
217 |
+
# UI処理 ========================================
|
218 |
+
|
219 |
+
# ユーザー名表示
|
220 |
+
def hello(profile: gr.OAuthProfile | None) -> Tuple[str, str]:
|
221 |
+
if profile is None:
|
222 |
+
return "プライベートデータセット取得のためにログインしてください。", who.value
|
223 |
+
who.value = profile.username
|
224 |
+
return f'{profile.username} さん、よろしくお願いいたします。', who.value
|
225 |
+
|
226 |
+
|
227 |
+
# テーマの状態
|
228 |
+
theme_ = gr.themes.Default()
|
229 |
+
|
230 |
+
# 後のCSSデザイン変更用
|
231 |
+
|
232 |
+
|
233 |
+
def load_css():
|
234 |
+
with open("style.css", "r") as file:
|
235 |
+
css_content = file.read()
|
236 |
+
return css_content
|
237 |
+
|
238 |
+
|
239 |
+
# Gradio 画面 ============================================
|
240 |
+
with gr.Blocks(theme=theme_, css=load_css()) as demo:
|
241 |
+
|
242 |
+
gr.Markdown("# データセット アノテーション for Tanuki (Phase2)")
|
243 |
+
|
244 |
+
with gr.Tab("アノテーション (シングルターン)"):
|
245 |
+
|
246 |
+
with gr.Row(equal_height=True):
|
247 |
+
|
248 |
+
gr.LoginButton(value="HuggingFace ログイン",
|
249 |
+
logout_value="HuggingFace ログアウト", scale=1)
|
250 |
+
|
251 |
+
# ユーザー名
|
252 |
+
gr_profile_name = gr.Markdown()
|
253 |
+
demo.load(hello, inputs=None, outputs=[gr_profile_name, who])
|
254 |
+
|
255 |
+
with gr.Row():
|
256 |
+
|
257 |
+
def dropdown_select(select_value) -> None:
|
258 |
+
select_dropdown_dataset.value = select_value
|
259 |
+
|
260 |
+
# 対象データセットの選択
|
261 |
+
gr_dropdown_dataset = gr.Dropdown(
|
262 |
+
label="データセット選択 ①",
|
263 |
+
choices=dropdown_dataset_list.value,
|
264 |
+
value=select_dropdown_dataset.value,
|
265 |
+
elem_id="dataset_sel",
|
266 |
+
scale=2)
|
267 |
+
|
268 |
+
gr_dropdown_dataset.change(
|
269 |
+
dropdown_select,
|
270 |
+
inputs=[gr_dropdown_dataset]
|
271 |
+
)
|
272 |
+
|
273 |
+
gr_data_load_btn = gr.Button("② データセットを読み込む")
|
274 |
+
|
275 |
+
with gr.Column() as content_column:
|
276 |
+
with gr.Tab("③ シンプル(良・悪)"):
|
277 |
+
with gr.Column():
|
278 |
+
with gr.Row(equal_height=True):
|
279 |
+
good_btn = gr.Button("良い", interactive=False)
|
280 |
+
bad_btn = gr.Button("悪い", interactive=False)
|
281 |
+
|
282 |
+
gr_question_text_1 = gr.Textbox(
|
283 |
+
label="質問: ", lines=5, interactive=False)
|
284 |
+
|
285 |
+
gr_answer_text_1 = gr.Textbox(
|
286 |
+
label="回答: 訂正頂けると品質が��がります。",
|
287 |
+
lines=20,
|
288 |
+
interactive=True)
|
289 |
+
|
290 |
+
with gr.Tab("③ 5段階評価"):
|
291 |
+
|
292 |
+
gr_question_text_3 = gr.Textbox(
|
293 |
+
label="質問: ", lines=5, interactive=False)
|
294 |
+
|
295 |
+
with gr.Row() as score_btn:
|
296 |
+
gr_score_1 = gr.Button("1: 低品質", interactive=False)
|
297 |
+
gr_score_2 = gr.Button("2: 悪い", interactive=False)
|
298 |
+
gr_score_3 = gr.Button("3: 普通", interactive=False)
|
299 |
+
gr_score_4 = gr.Button("4: 良い", interactive=False)
|
300 |
+
gr_score_5 = gr.Button("5: 高品質", interactive=False)
|
301 |
+
|
302 |
+
gr_answer_text_3 = gr.Textbox(
|
303 |
+
label="回答: 訂正して頂けると品質が上がります。", lines=20, interactive=True)
|
304 |
+
|
305 |
+
# 5段階評価ボタンのクリックイベントを定義
|
306 |
+
def score_button_clicked(button_value):
|
307 |
+
good.value = False
|
308 |
+
bad.value = False
|
309 |
+
score.value = button_value
|
310 |
+
|
311 |
+
gr_data_load_btn.click(
|
312 |
+
dataset_load_fn,
|
313 |
+
inputs=None,
|
314 |
+
outputs=[gr_question_text_1,
|
315 |
+
gr_answer_text_1,
|
316 |
+
gr_question_text_3,
|
317 |
+
gr_answer_text_3,
|
318 |
+
good_btn,
|
319 |
+
bad_btn,
|
320 |
+
gr_score_1,
|
321 |
+
gr_score_2,
|
322 |
+
gr_score_3,
|
323 |
+
gr_score_4,
|
324 |
+
gr_score_5,
|
325 |
+
]
|
326 |
+
)
|
327 |
+
|
328 |
+
def update_annotation(
|
329 |
+
input_ans_1: str = None,
|
330 |
+
input_ans_3: str = None,
|
331 |
+
is_good: bool = None, # good/bad を表すフラグを追加
|
332 |
+
score_value: int = None # 5段階評価の値、good/badの場合はNone
|
333 |
+
) -> Tuple[gr.update, gr.update, gr.update, gr.update]:
|
334 |
+
|
335 |
+
# good/bad と score の状態を更新
|
336 |
+
if score_value is not None: # 5段階評価の場合
|
337 |
+
good.value = False
|
338 |
+
bad.value = False
|
339 |
+
score.value = score_value
|
340 |
+
else: # good/bad評価の場合
|
341 |
+
good.value = is_good
|
342 |
+
bad.value = not is_good
|
343 |
+
|
344 |
+
# 変更を検知 (5段階評価の場合も処理するように変更)
|
345 |
+
if input_ans_1 is not None and initial_answer_text_1.value != input_ans_1:
|
346 |
+
is_proofreading_1.value = True
|
347 |
+
answer_text_1.value = input_ans_1
|
348 |
+
else:
|
349 |
+
answer_text_1.value = ""
|
350 |
+
|
351 |
+
if input_ans_3 is not None and initial_answer_text_2.value != input_ans_3:
|
352 |
+
is_proofreading_2.value = True
|
353 |
+
answer_text_2.value = input_ans_3
|
354 |
+
else:
|
355 |
+
answer_text_2.value = ""
|
356 |
+
|
357 |
+
# 表示更新
|
358 |
+
# indexを進める
|
359 |
+
select_idx.value += 1
|
360 |
+
|
361 |
+
df = select_dataset.value["train"]
|
362 |
+
|
363 |
+
# ループさせるか、エラー処理を行う
|
364 |
+
if select_idx.value >= len(df):
|
365 |
+
select_idx.value = 0
|
366 |
+
|
367 |
+
# データセットに追加
|
368 |
+
# 元のindex番号(dataset_id)を指定して保存
|
369 |
+
save_annotation(
|
370 |
+
select_dropdown_dataset.value,
|
371 |
+
# datasetIdは元のindex番号を使用
|
372 |
+
df.iloc[select_idx.value]['index'],
|
373 |
+
who.value,
|
374 |
+
good.value,
|
375 |
+
bad.value,
|
376 |
+
score.value,
|
377 |
+
is_proofreading_1.value,
|
378 |
+
answer_text_1.value,
|
379 |
+
is_proofreading_2.value,
|
380 |
+
answer_text_2.value
|
381 |
+
)
|
382 |
+
|
383 |
+
# Nextデータ初期化
|
384 |
+
is_proofreading_1.value = False
|
385 |
+
is_proofreading_2.value = False
|
386 |
+
initial_answer_text_1.value = df.iloc[select_idx.value]["answer"]
|
387 |
+
initial_answer_text_2.value = df.iloc[select_idx.value]["answer"]
|
388 |
+
|
389 |
+
return gr.update(value=df.iloc[select_idx.value]["question"]), \
|
390 |
+
gr.update(value=df.iloc[select_idx.value]["answer"]), \
|
391 |
+
gr.update(value=df.iloc[select_idx.value]["question"]), \
|
392 |
+
gr.update(value=df.iloc[select_idx.value]["answer"])
|
393 |
+
|
394 |
+
def good_click(input_ans_1, input_ans_3):
|
395 |
+
return update_annotation(input_ans_1=input_ans_1, input_ans_3=input_ans_3, is_good=True)
|
396 |
+
|
397 |
+
good_btn.click(
|
398 |
+
good_click,
|
399 |
+
inputs=[
|
400 |
+
gr_answer_text_1,
|
401 |
+
gr_answer_text_3
|
402 |
+
],
|
403 |
+
outputs=[gr_question_text_1,
|
404 |
+
gr_answer_text_1,
|
405 |
+
gr_question_text_3,
|
406 |
+
gr_answer_text_3]
|
407 |
+
)
|
408 |
+
|
409 |
+
def bad_click(input_ans_1, input_ans_3):
|
410 |
+
return update_annotation(input_ans_1=input_ans_1, input_ans_3=input_ans_3, is_good=False)
|
411 |
+
|
412 |
+
bad_btn.click(
|
413 |
+
bad_click,
|
414 |
+
inputs=[
|
415 |
+
gr_answer_text_1,
|
416 |
+
gr_answer_text_3
|
417 |
+
],
|
418 |
+
outputs=[gr_question_text_1,
|
419 |
+
gr_answer_text_1,
|
420 |
+
gr_question_text_3,
|
421 |
+
gr_answer_text_3]
|
422 |
+
)
|
423 |
+
|
424 |
+
# 5段階評価ボタンのクリックイベント
|
425 |
+
gr_score_1.click(lambda x: update_annotation(input_ans_1=x, input_ans_3=x, score_value=1),
|
426 |
+
inputs=[gr_answer_text_3], outputs=[gr_question_text_1, gr_answer_text_1, gr_question_text_3, gr_answer_text_3])
|
427 |
+
gr_score_2.click(lambda x: update_annotation(input_ans_1=x, input_ans_3=x, score_value=2),
|
428 |
+
inputs=[gr_answer_text_3], outputs=[gr_question_text_1, gr_answer_text_1, gr_question_text_3, gr_answer_text_3])
|
429 |
+
gr_score_3.click(lambda x: update_annotation(input_ans_1=x, input_ans_3=x, score_value=3),
|
430 |
+
inputs=[gr_answer_text_3], outputs=[gr_question_text_1, gr_answer_text_1, gr_question_text_3, gr_answer_text_3])
|
431 |
+
gr_score_4.click(lambda x: update_annotation(input_ans_1=x, input_ans_3=x, score_value=4),
|
432 |
+
inputs=[gr_answer_text_3], outputs=[gr_question_text_1, gr_answer_text_1, gr_question_text_3, gr_answer_text_3])
|
433 |
+
gr_score_5.click(lambda x: update_annotation(input_ans_1=x, input_ans_3=x, score_value=5),
|
434 |
+
inputs=[gr_answer_text_3], outputs=[gr_question_text_1, gr_answer_text_1, gr_question_text_3, gr_answer_text_3])
|
435 |
+
|
436 |
+
# TODO Tab切り替えで、アノテ済みの一覧を表示する
|
437 |
+
# with gr.Tab("アノテ済みデータセット(管理画面)"):
|
438 |
+
# タブを切り替えた時にデータ表示を更新する
|
439 |
+
|
440 |
+
if __name__ == "__main__":
|
441 |
+
demo.launch()
|
note.txt
CHANGED
@@ -5,4 +5,4 @@ Secretsに作成したTokenを
|
|
5 |
HF_TOKENに設定して頂けますでしょうか?
|
6 |
|
7 |
- team-hatakeyama-phase2/annotation_tanuki_phase2
|
8 |
-
側も設定が必要?
|
|
|
5 |
HF_TOKENに設定して頂けますでしょうか?
|
6 |
|
7 |
- team-hatakeyama-phase2/annotation_tanuki_phase2
|
8 |
+
側も設定が必要?
|
requirements.txt
CHANGED
@@ -1,4 +1,3 @@
|
|
1 |
huggingface_hub==0.22.2
|
2 |
-
|
3 |
-
transformers
|
4 |
datasets
|
|
|
1 |
huggingface_hub==0.22.2
|
2 |
+
gradio
|
|
|
3 |
datasets
|