Add LabelDescription list to labeling tasks, make +1/-1 emojis exclusive (#947)

* add LabelDescription list to labeling tasks

* make +1 & -1 emoji exclusive (only one of both or none)

* add red_flag emoji to message when reported

* fix task's valid labels

* fix typo

oasst-shared/oasst_shared/schemas/protocol.py

@@ -231,12 +231,20 @@ class LabelTaskDisposition(str, enum.Enum):
     spam = "spam"
 
 
+class LabelDescription(BaseModel):
+    name: str
+    widget: str
+    display_text: str
+    help_text: Optional[str]
+
+
 class AbstractLabelTask(Task):
     message_id: UUID
     valid_labels: list[str]
     mandatory_labels: Optional[list[str]]
     mode: Optional[LabelTaskMode]
     disposition: Optional[LabelTaskDisposition]
+    labels: Optional[list[LabelDescription]]
 
 
 class LabelInitialPromptTask(AbstractLabelTask):
@@ -324,39 +332,48 @@ class MessageRanking(Interaction):
     ranking: conlist(item_type=int, min_items=1)
 
 
+class LabelWidget(str, enum.Enum):
+    yes_no = "yes_no"
+    flag = "flag"
+    likert = "likert"
+
+
 class TextLabel(str, enum.Enum):
     """A label for a piece of text."""
 
-    def __new__(cls, label: str, display_text: str = "", help_text: str = None):
+    def __new__(cls, label: str, widget: LabelWidget, display_text: str = "", help_text: str = None):
         obj = str.__new__(cls, label)
         obj._value_ = label
+        obj.widget = widget
         obj.display_text = display_text
         obj.help_text = help_text
         return obj
 
-    spam = "spam", "Seems to be intentionally low-quality or irrelevant"
-    fails_task = "fails_task", "Fails to follow the correct instruction / task"
-    not_appropriate = "not_appropriate", "Inappropriate for customer assistant"
-    violence = "violence", "Encourages or fails to discourage violence/abuse/terrorism/self-harm"
-    excessive_harm = (
-        "excessive_harm",
-        "Content likely to cause excessive harm not justifiable in the context",
-        "Harm refers to physical or mental damage or injury to someone or something. Excessive refers to a reasonable threshold of harm in the context, for instance damaging skin is not excessive in the context of surgery.",
-    )
-    sexual_content = "sexual_content", "Contains sexual content"
-    toxicity = "toxicity", "Contains rude, abusive, profane or insulting content"
-    moral_judgement = "moral_judgement", "Expresses moral judgement"
-    political_content = "political_content", "Expresses political views"
-    humor = "humor", "Contains humorous content including sarcasm"
+    # yes/no questions
+    spam = "spam", LabelWidget.yes_no, "Seems to be intentionally low-quality or irrelevant"
+    fails_task = "fails_task", LabelWidget.yes_no, "Fails to follow the correct instruction / task"
+
+    # flags
+    pii = "pii", LabelWidget.flag, "Contains personal identifiable information (PII)"
+    not_appropriate = "not_appropriate", LabelWidget.flag, "Inappropriate"
     hate_speech = (
         "hate_speech",
+        LabelWidget.flag,
         "Content is abusive or threatening and expresses prejudice against a protected characteristic",
-        "Prejudice refers to preconceived views not based on reason. Protected characteristics include gender, ethnicity, religion, sexual orientation, and similar characteristics.",
+        "Prejudice refers to preconceived views not based on reason. Protected characteristics "
+        "include gender, ethnicity, religion, sexual orientation, and similar characteristics.",
     )
-    threat = "threat", "Contains a threat against a person or persons"
-    misleading = "misleading", "Contains text which is incorrect or misleading"
-    helpful = "helpful", "Completes the task to a high standard"
-    creative = "creative", "Expresses creativity in responding to the task"
+    sexual_content = "sexual_content", LabelWidget.flag, "Contains sexual content"
+    moral_judgement = "moral_judgement", LabelWidget.flag, "Expresses moral judgement"
+    political_content = "political_content", LabelWidget.flag, "Expresses political views"
+
+    # likert
+    quality = "quality", LabelWidget.likert, "Overall subjective quality rating of the message"
+    toxicity = "toxicity", LabelWidget.likert, "Rude, abusive, profane or insulting content"
+    humor = "humor", LabelWidget.likert, "Humorous content including sarcasm"
+    helpfulness = "helpfulness", LabelWidget.likert, "Helpfulness of the message"
+    creativity = "creativity", LabelWidget.likert, "Creativity"
+    violence = "violence", LabelWidget.likert, "Violence/abuse/terrorism/self-harm"
 
 
 class TextLabels(Interaction):
@@ -367,6 +384,7 @@ class TextLabels(Interaction):
     labels: dict[TextLabel, float]
     message_id: UUID
     task_id: Optional[UUID]
+    is_report: Optional[bool]
 
     @property
     def has_message_id(self) -> bool: