add api/v1/text_labels/valid_labels backend endpoint (#498)

Provides the set of valid text_lables with descriptions via a REST endpoint.

backend/oasst_backend/api/v1/text_labels.py

@@ -3,6 +3,7 @@ from fastapi.security.api_key import APIKey
 from loguru import logger
 from oasst_backend.api import deps
 from oasst_backend.prompt_repository import PromptRepository
+from oasst_backend.schemas.text_labels import LabelOption, ValidLabelsResponse
 from oasst_shared.schemas import protocol as protocol_schema
 from sqlmodel import Session
 from starlette.status import HTTP_204_NO_CONTENT, HTTP_400_BAD_REQUEST
@@ -32,3 +33,10 @@ def label_text(
         raise HTTPException(
             status_code=HTTP_400_BAD_REQUEST,
         )
+
+
+@router.get("/valid_labels")
+def get_valid_lables() -> ValidLabelsResponse:
+    return ValidLabelsResponse(
+        valid_labels=[LabelOption(name=l.value, description=l.description) for l in protocol_schema.TextLabel]
+    )

backend/oasst_backend/schemas/text_labels.py

@@ -0,0 +1,10 @@
+from pydantic import BaseModel
+
+
+class LabelOption(BaseModel):
+    name: str
+    description: str
+
+
+class ValidLabelsResponse(BaseModel):
+    valid_labels: list[LabelOption]

oasst-shared/oasst_shared/schemas/protocol.py

@@ -265,10 +265,23 @@ class MessageRanking(Interaction):
 class TextLabel(str, enum.Enum):
     """A label for a piece of text."""
 
+    def __new__(cls, label, description=""):
+        obj = str.__new__(cls, label)
+        obj._value_ = label
+        obj.description = description
+        return obj
+
     spam = "spam"
-    violence = "violence"
-    sexual_content = "sexual_content"
+    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"
+    harmful = (
+        "harmful",
+        "The advice given in the output is harmful or counter-productive. This may be in addition to, but is distinct from the question about encouraging violence/abuse/terrorism/self-harm.",
+    )
+    sexual_content = "sexual_content", "Contains sexual content"
     toxicity = "toxicity"
+    moral_judgement = "moral_judgement", "Expresses moral judgement"
     political_content = "political_content"
     humor = "humor"
     sarcasm = "sarcasm"