readme修正 ５段階評価button並び順修正

.gitignore

@@ -162,5 +162,7 @@ cython_debug/
 #.idea/
 
 user_annotation/*
-
-backup.py
+run.py
+backup.py
+huggingface_dataset.txt
+note.txt

README.md

@@ -8,6 +8,18 @@ app_file: app.py
 pinned: false
 license: unknown
 
+language:
+- ja
+
+datasets:
+- team-hatakeyama-phase2/annotation_tanuki_phase2
+- hatakeyama-llm-team/AutoGeneratedJapaneseQA
+- hatakeyama-llm-team/AutoGeneratedJapaneseQA-other
+- kanhatakeyama/ChatbotArenaJaMixtral8x22b
+- kanhatakeyama/OrcaJaMixtral8x22b
+- kanhatakeyama/LogicalDatasetsByMixtral8x22b
+- kanhatakeyama/AutoMultiTurnByMixtral8x22b
+
 hf_oauth: true
 # optional, default duration is 8 hours/480 minutes. Max duration is 30 days/43200 minutes.
 hf_oauth_expiration_minutes: 480

app.py

@@ -104,6 +104,7 @@ answer_text_2 = gr.State("")  # answer_2 回答
 # 未整理
 # データ読み込み　========================================
 
+
 def dataset_load_fn() -> Tuple[
         str,
         str,
@@ -228,6 +229,8 @@ def save_annotation(
 # UI処理 ========================================
 
 # ユーザー名表示
+
+
 def hello(profile: gr.OAuthProfile | None) -> Tuple[str, str]:
     if profile is None:
         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
@@ -289,7 +292,8 @@ with gr.Blocks(theme=theme_, css=load_css()) as demo:
                 with gr.Column():
                     with gr.Row(equal_height=True):
                         good_btn = gr.Button("良い", interactive=False)
-                        unknown_btn = gr.Button("分からない", interactive=False)  # 「分からない」ボタンを追加
+                        unknown_btn = gr.Button(
+                            "分からない", interactive=False)  # 「分からない」ボタンを追加
                         bad_btn = gr.Button("悪い", interactive=False)
 
                 gr_question_text_1 = gr.Textbox(
@@ -306,11 +310,11 @@ with gr.Blocks(theme=theme_, css=load_css()) as demo:
                     label="質問: ", lines=5, interactive=False)
 
                 with gr.Row() as score_btn:
-                    gr_score_1 = gr.Button("1: 低品質", interactive=False)
-                    gr_score_2 = gr.Button("2: 悪い", interactive=False)
-                    gr_score_3 = gr.Button("3: 普通", interactive=False)
-                    gr_score_4 = gr.Button("4: 良い", interactive=False)
                     gr_score_5 = gr.Button("5: 高品質", interactive=False)
+                    gr_score_4 = gr.Button("4: 良い", interactive=False)
+                    gr_score_3 = gr.Button("3: 普通", interactive=False)
+                    gr_score_2 = gr.Button("2: 悪い", interactive=False)
+                    gr_score_1 = gr.Button("1: 低品質", interactive=False)
 
                 gr_answer_text_3 = gr.Textbox(
                     label="回答: 訂正して頂けると品質が上がります。", lines=20, interactive=True)
@@ -319,7 +323,7 @@ with gr.Blocks(theme=theme_, css=load_css()) as demo:
             def score_button_clicked(button_value):
                 good.value = False
                 bad.value = False
-                unknown_quality.value = False # 「分からない」の状態をリセット
+                unknown_quality.value = False  # 「分からない」の状態をリセット
                 score.value = button_value
 
             gr_data_load_btn.click(
@@ -349,6 +353,34 @@ with gr.Blocks(theme=theme_, css=load_css()) as demo:
             ) -> Tuple[gr.update, gr.update, gr.update, gr.update]:
 
                 # good/bad/unknown と score の状態を更新
+                update_evaluation_state(is_good, is_unknown, score_value)
+
+                # 変更を検知 (5段階評価の場合も処理するように変更)
+                update_answer_state(input_ans_1, input_ans_3)
+
+                # 表示更新
+                # indexを進める
+                select_idx.value += 1
+
+                df = select_dataset.value["train"]
+
+                # ループさせるか、エラー処理を行う
+                if select_idx.value >= len(df):
+                    select_idx.value = 0
+
+                # データセットに追加
+                # 元のindex番号(dataset_id)を指定して保存
+                save_annotation_data(df)
+
+                # Nextデータ初期化
+                initialize_next_data(df)
+
+                return gr.update(value=df.iloc[select_idx.value]["question"]), \
+                    gr.update(value=df.iloc[select_idx.value]["answer"]), \
+                    gr.update(value=df.iloc[select_idx.value]["question"]), \
+                    gr.update(value=df.iloc[select_idx.value]["answer"])
+
+            def update_evaluation_state(is_good, is_unknown, score_value):
                 if score_value is not None:  # 5段階評価の場合
                     good.value = False
                     bad.value = False
@@ -359,7 +391,7 @@ with gr.Blocks(theme=theme_, css=load_css()) as demo:
                     bad.value = not is_good if not is_unknown else False
                     unknown_quality.value = is_unknown  # 「分からない」の状態を設定
 
-                # 変更を検知 (5段階評価の場合も処理するように変更)
+            def update_answer_state(input_ans_1, input_ans_3):
                 if input_ans_1 is not None and initial_answer_text_1.value != input_ans_1:
                     is_proofreading_1.value = True
                     answer_text_1.value = input_ans_1
@@ -372,18 +404,7 @@ with gr.Blocks(theme=theme_, css=load_css()) as demo:
                 else:
                     answer_text_2.value = ""
 
-                # 表示更新
-                # indexを進める
-                select_idx.value += 1
-
-                df = select_dataset.value["train"]
-
-                # ループさせるか、エラー処理を行う
-                if select_idx.value >= len(df):
-                    select_idx.value = 0
-
-                # データセットに追加
-                # 元のindex番号(dataset_id)を指定して保存
+            def save_annotation_data(df):
                 save_annotation(
                     select_dropdown_dataset.value,
                     # datasetIdは元のindex番号を使用
@@ -399,17 +420,12 @@ with gr.Blocks(theme=theme_, css=load_css()) as demo:
                     answer_text_2.value
                 )
 
-                # Nextデータ初期化
+            def initialize_next_data(df):
                 is_proofreading_1.value = False
                 is_proofreading_2.value = False
                 initial_answer_text_1.value = df.iloc[select_idx.value]["answer"]
                 initial_answer_text_2.value = df.iloc[select_idx.value]["answer"]
 
-                return gr.update(value=df.iloc[select_idx.value]["question"]), \
-                    gr.update(value=df.iloc[select_idx.value]["answer"]), \
-                    gr.update(value=df.iloc[select_idx.value]["question"]), \
-                    gr.update(value=df.iloc[select_idx.value]["answer"])
-
             def good_click(input_ans_1, input_ans_3):
                 return update_annotation(input_ans_1=input_ans_1, input_ans_3=input_ans_3, is_good=True, is_unknown=False)
 
@@ -458,15 +474,15 @@ with gr.Blocks(theme=theme_, css=load_css()) as demo:
             )
 
             # 5段階評価ボタンのクリックイベント
-            gr_score_1.click(lambda x: update_annotation(input_ans_1=x, input_ans_3=x, is_unknown=False, score_value=1),
+            gr_score_1.click(lambda x: update_annotation(input_ans_1=x, is_unknown=False, score_value=1),
                              inputs=[gr_answer_text_3], outputs=[gr_question_text_1, gr_answer_text_1, gr_question_text_3, gr_answer_text_3])
-            gr_score_2.click(lambda x: update_annotation(input_ans_1=x, input_ans_3=x, is_unknown=False, score_value=2),
+            gr_score_2.click(lambda x: update_annotation(input_ans_1=x, is_unknown=False, score_value=2),
                              inputs=[gr_answer_text_3], outputs=[gr_question_text_1, gr_answer_text_1, gr_question_text_3, gr_answer_text_3])
-            gr_score_3.click(lambda x: update_annotation(input_ans_1=x, input_ans_3=x, is_unknown=False, score_value=3),
+            gr_score_3.click(lambda x: update_annotation(input_ans_1=x, is_unknown=False, score_value=3),
                              inputs=[gr_answer_text_3], outputs=[gr_question_text_1, gr_answer_text_1, gr_question_text_3, gr_answer_text_3])
-            gr_score_4.click(lambda x: update_annotation(input_ans_1=x, input_ans_3=x, is_unknown=False, score_value=4),
+            gr_score_4.click(lambda x: update_annotation(input_ans_1=x, is_unknown=False, score_value=4),
                              inputs=[gr_answer_text_3],  outputs=[gr_question_text_1, gr_answer_text_1, gr_question_text_3, gr_answer_text_3])
-            gr_score_5.click(lambda x: update_annotation(input_ans_1=x, input_ans_3=x, is_unknown=False, score_value=5),
+            gr_score_5.click(lambda x: update_annotation(input_ans_1=x, is_unknown=False, score_value=5),
                              inputs=[gr_answer_text_3],  outputs=[gr_question_text_1, gr_answer_text_1, gr_question_text_3, gr_answer_text_3])
 
             # TODO Tab切り替えで、アノテ済みの一覧を表示する

huggingface_dataset.txt

@@ -59,6 +59,13 @@ fileSize 116MB
 - question string
 - answer string
 
+# susumuota/SyntheticTextWikiTranslate-askllm-v1
+- mode string
+- url string
+- ja string
+- eng string
+- askllm_score float64
+
 # kevineen/Tanuki-Phase2-annotation-dataset
 - index int64
 - datetime string
@@ -72,11 +79,4 @@ fileSize 116MB
 - is_proofreading_1 bool
 - answer_text_1 string
 - is_proofreading_2 bool
-- answer_text_2 string
-
-# susumuota/SyntheticTextWikiTranslate-askllm-v1
-- mode string
-- url string
-- ja string
-- eng string
-- askllm_score float64
+- answer_text_2 string

run.py

@@ -1,3 +1,5 @@
+# gradio run.pyでHotReload開発
+
 import os
 from dotenv import load_dotenv
 
@@ -103,7 +105,6 @@ answer_text_2 = gr.State("")  # answer_2 回答
 
 # 未整理
 # データ読み込み　========================================
-
 def dataset_load_fn() -> Tuple[
         str,
         str,
@@ -163,7 +164,7 @@ annotation_file = Path("user_annotation/") / f"data_{uuid.uuid4()}.json"
 annotated_folder = annotation_file.parent
 
 scheduler = CommitScheduler(
-    repo_id=output_dataset[0],
+    repo_id=output_dataset[1],
     repo_type="dataset",
     folder_path=annotated_folder,
     path_in_repo="data",  # Spaceの場合の保存先フォルダー
@@ -172,8 +173,6 @@ scheduler = CommitScheduler(
 )
 
 # CommitScheduler (HFへのデータアップロード
-
-
 def save_annotation(
         dataset_name: str,
         dataset_id: int,
@@ -230,6 +229,8 @@ def save_annotation(
 # UI処理 ========================================
 
 # ユーザー名表示
+
+
 def hello(profile: gr.OAuthProfile | None) -> Tuple[str, str]:
     if profile is None:
         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
@@ -291,7 +292,8 @@ with gr.Blocks(theme=theme_, css=load_css()) as demo:
                 with gr.Column():
                     with gr.Row(equal_height=True):
                         good_btn = gr.Button("良い", interactive=False)
-                        unknown_btn = gr.Button("分からない", interactive=False)  # 「分からない」ボタンを追加
+                        unknown_btn = gr.Button(
+                            "分からない", interactive=False)  # 「分からない」ボタンを追加
                         bad_btn = gr.Button("悪い", interactive=False)
 
                 gr_question_text_1 = gr.Textbox(
@@ -308,11 +310,11 @@ with gr.Blocks(theme=theme_, css=load_css()) as demo:
                     label="質問: ", lines=5, interactive=False)
 
                 with gr.Row() as score_btn:
-                    gr_score_1 = gr.Button("1: 低品質", interactive=False)
-                    gr_score_2 = gr.Button("2: 悪い", interactive=False)
-                    gr_score_3 = gr.Button("3: 普通", interactive=False)
-                    gr_score_4 = gr.Button("4: 良い", interactive=False)
                     gr_score_5 = gr.Button("5: 高品質", interactive=False)
+                    gr_score_4 = gr.Button("4: 良い", interactive=False)
+                    gr_score_3 = gr.Button("3: 普通", interactive=False)
+                    gr_score_2 = gr.Button("2: 悪い", interactive=False)
+                    gr_score_1 = gr.Button("1: 低品質", interactive=False)
 
                 gr_answer_text_3 = gr.Textbox(
                     label="回答: 訂正して頂けると品質が上がります。", lines=20, interactive=True)
@@ -321,7 +323,7 @@ with gr.Blocks(theme=theme_, css=load_css()) as demo:
             def score_button_clicked(button_value):
                 good.value = False
                 bad.value = False
-                unknown_quality.value = False # 「分からない」の状態をリセット
+                unknown_quality.value = False  # 「分からない」の状態をリセット
                 score.value = button_value
 
             gr_data_load_btn.click(
@@ -351,6 +353,34 @@ with gr.Blocks(theme=theme_, css=load_css()) as demo:
             ) -> Tuple[gr.update, gr.update, gr.update, gr.update]:
 
                 # good/bad/unknown と score の状態を更新
+                update_evaluation_state(is_good, is_unknown, score_value)
+
+                # 変更を検知 (5段階評価の場合も処理するように変更)
+                update_answer_state(input_ans_1, input_ans_3)
+
+                # 表示更新
+                # indexを進める
+                select_idx.value += 1
+
+                df = select_dataset.value["train"]
+
+                # ループさせるか、エラー処理を行う
+                if select_idx.value >= len(df):
+                    select_idx.value = 0
+
+                # データセットに追加
+                # 元のindex番号(dataset_id)を指定して保存
+                save_annotation_data(df)
+
+                # Nextデータ初期化
+                initialize_next_data(df)
+
+                return gr.update(value=df.iloc[select_idx.value]["question"]), \
+                    gr.update(value=df.iloc[select_idx.value]["answer"]), \
+                    gr.update(value=df.iloc[select_idx.value]["question"]), \
+                    gr.update(value=df.iloc[select_idx.value]["answer"])
+
+            def update_evaluation_state(is_good, is_unknown, score_value):
                 if score_value is not None:  # 5段階評価の場合
                     good.value = False
                     bad.value = False
@@ -361,7 +391,7 @@ with gr.Blocks(theme=theme_, css=load_css()) as demo:
                     bad.value = not is_good if not is_unknown else False
                     unknown_quality.value = is_unknown  # 「分からない」の状態を設定
 
-                # 変更を検知 (5段階評価の場合も処理するように変更)
+            def update_answer_state(input_ans_1, input_ans_3):
                 if input_ans_1 is not None and initial_answer_text_1.value != input_ans_1:
                     is_proofreading_1.value = True
                     answer_text_1.value = input_ans_1
@@ -374,18 +404,7 @@ with gr.Blocks(theme=theme_, css=load_css()) as demo:
                 else:
                     answer_text_2.value = ""
 
-                # 表示更新
-                # indexを進める
-                select_idx.value += 1
-
-                df = select_dataset.value["train"]
-
-                # ループさせるか、エラー処理を行う
-                if select_idx.value >= len(df):
-                    select_idx.value = 0
-
-                # データセットに追加
-                # 元のindex番号(dataset_id)を指定して保存
+            def save_annotation_data(df):
                 save_annotation(
                     select_dropdown_dataset.value,
                     # datasetIdは元のindex番号を使用
@@ -401,17 +420,12 @@ with gr.Blocks(theme=theme_, css=load_css()) as demo:
                     answer_text_2.value
                 )
 
-                # Nextデータ初期化
+            def initialize_next_data(df):
                 is_proofreading_1.value = False
                 is_proofreading_2.value = False
                 initial_answer_text_1.value = df.iloc[select_idx.value]["answer"]
                 initial_answer_text_2.value = df.iloc[select_idx.value]["answer"]
 
-                return gr.update(value=df.iloc[select_idx.value]["question"]), \
-                    gr.update(value=df.iloc[select_idx.value]["answer"]), \
-                    gr.update(value=df.iloc[select_idx.value]["question"]), \
-                    gr.update(value=df.iloc[select_idx.value]["answer"])
-
             def good_click(input_ans_1, input_ans_3):
                 return update_annotation(input_ans_1=input_ans_1, input_ans_3=input_ans_3, is_good=True, is_unknown=False)
 
@@ -460,15 +474,15 @@ with gr.Blocks(theme=theme_, css=load_css()) as demo:
             )
 
             # 5段階評価ボタンのクリックイベント
-            gr_score_1.click(lambda x: update_annotation(input_ans_1=x, input_ans_3=x, is_unknown=False, score_value=1),
+            gr_score_1.click(lambda x: update_annotation(input_ans_1=x, is_unknown=False, score_value=1),
                              inputs=[gr_answer_text_3], outputs=[gr_question_text_1, gr_answer_text_1, gr_question_text_3, gr_answer_text_3])
-            gr_score_2.click(lambda x: update_annotation(input_ans_1=x, input_ans_3=x, is_unknown=False, score_value=2),
+            gr_score_2.click(lambda x: update_annotation(input_ans_1=x, is_unknown=False, score_value=2),
                              inputs=[gr_answer_text_3], outputs=[gr_question_text_1, gr_answer_text_1, gr_question_text_3, gr_answer_text_3])
-            gr_score_3.click(lambda x: update_annotation(input_ans_1=x, input_ans_3=x, is_unknown=False, score_value=3),
+            gr_score_3.click(lambda x: update_annotation(input_ans_1=x, is_unknown=False, score_value=3),
                              inputs=[gr_answer_text_3], outputs=[gr_question_text_1, gr_answer_text_1, gr_question_text_3, gr_answer_text_3])
-            gr_score_4.click(lambda x: update_annotation(input_ans_1=x, input_ans_3=x, is_unknown=False, score_value=4),
+            gr_score_4.click(lambda x: update_annotation(input_ans_1=x, is_unknown=False, score_value=4),
                              inputs=[gr_answer_text_3],  outputs=[gr_question_text_1, gr_answer_text_1, gr_question_text_3, gr_answer_text_3])
-            gr_score_5.click(lambda x: update_annotation(input_ans_1=x, input_ans_3=x, is_unknown=False, score_value=5),
+            gr_score_5.click(lambda x: update_annotation(input_ans_1=x, is_unknown=False, score_value=5),
                              inputs=[gr_answer_text_3],  outputs=[gr_question_text_1, gr_answer_text_1, gr_question_text_3, gr_answer_text_3])
 
             # TODO Tab切り替えで、アノテ済みの一覧を表示する