Spaces:
Sleeping
Sleeping
Commit ·
c651e61
1
Parent(s): f15e9b2
refactor(models): clean up imports and add module docstring
Browse files
models.py
CHANGED
|
@@ -1,5 +1,11 @@
|
|
| 1 |
-
|
|
|
|
|
|
|
|
|
|
| 2 |
from pydantic import BaseModel, Field, field_validator
|
|
|
|
|
|
|
|
|
|
| 3 |
|
| 4 |
class StrategyType(str, Enum):
|
| 5 |
ROLEPLAY = "roleplay"
|
|
@@ -10,6 +16,7 @@ class StrategyType(str, Enum):
|
|
| 10 |
ENCODING = "encoding"
|
| 11 |
MULTI_TURN = "multi_turn"
|
| 12 |
|
|
|
|
| 13 |
class TargetCategory(str, Enum):
|
| 14 |
PRIVACY = "privacy"
|
| 15 |
MISINFORMATION = "misinformation"
|
|
@@ -17,6 +24,7 @@ class TargetCategory(str, Enum):
|
|
| 17 |
MANIPULATION = "manipulation"
|
| 18 |
ILLEGAL_ACTIVITY = "illegal_activity"
|
| 19 |
|
|
|
|
| 20 |
class AttackAction(BaseModel):
|
| 21 |
strategy_type: StrategyType
|
| 22 |
target_category: TargetCategory
|
|
@@ -31,6 +39,7 @@ class AttackAction(BaseModel):
|
|
| 31 |
v = v.replace(pattern, "[FILTERED]")
|
| 32 |
return v.strip()
|
| 33 |
|
|
|
|
| 34 |
class RedTeamObservation(BaseModel):
|
| 35 |
defender_response: str
|
| 36 |
defense_score: float = Field(ge=0.0, le=1.0)
|
|
@@ -41,6 +50,7 @@ class RedTeamObservation(BaseModel):
|
|
| 41 |
feedback: str
|
| 42 |
episode_id: str
|
| 43 |
|
|
|
|
| 44 |
class EpisodeState(BaseModel):
|
| 45 |
episode_id: str
|
| 46 |
turn: int
|
|
@@ -48,10 +58,12 @@ class EpisodeState(BaseModel):
|
|
| 48 |
attacks_so_far:int
|
| 49 |
is_active: bool
|
| 50 |
|
|
|
|
| 51 |
class StepResult(BaseModel):
|
| 52 |
observation: RedTeamObservation
|
| 53 |
reward: float
|
| 54 |
|
|
|
|
| 55 |
class ResetResponse(BaseModel):
|
| 56 |
observation: RedTeamObservation
|
| 57 |
episode_id: str
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
Shared Pydantic models and Enums for the RedTeamOS environment.
|
| 3 |
+
These define the API contract between the server, environment, and clients.
|
| 4 |
+
"""
|
| 5 |
from pydantic import BaseModel, Field, field_validator
|
| 6 |
+
from typing import Optional
|
| 7 |
+
from enum import Enum
|
| 8 |
+
|
| 9 |
|
| 10 |
class StrategyType(str, Enum):
|
| 11 |
ROLEPLAY = "roleplay"
|
|
|
|
| 16 |
ENCODING = "encoding"
|
| 17 |
MULTI_TURN = "multi_turn"
|
| 18 |
|
| 19 |
+
|
| 20 |
class TargetCategory(str, Enum):
|
| 21 |
PRIVACY = "privacy"
|
| 22 |
MISINFORMATION = "misinformation"
|
|
|
|
| 24 |
MANIPULATION = "manipulation"
|
| 25 |
ILLEGAL_ACTIVITY = "illegal_activity"
|
| 26 |
|
| 27 |
+
|
| 28 |
class AttackAction(BaseModel):
|
| 29 |
strategy_type: StrategyType
|
| 30 |
target_category: TargetCategory
|
|
|
|
| 39 |
v = v.replace(pattern, "[FILTERED]")
|
| 40 |
return v.strip()
|
| 41 |
|
| 42 |
+
|
| 43 |
class RedTeamObservation(BaseModel):
|
| 44 |
defender_response: str
|
| 45 |
defense_score: float = Field(ge=0.0, le=1.0)
|
|
|
|
| 50 |
feedback: str
|
| 51 |
episode_id: str
|
| 52 |
|
| 53 |
+
|
| 54 |
class EpisodeState(BaseModel):
|
| 55 |
episode_id: str
|
| 56 |
turn: int
|
|
|
|
| 58 |
attacks_so_far:int
|
| 59 |
is_active: bool
|
| 60 |
|
| 61 |
+
|
| 62 |
class StepResult(BaseModel):
|
| 63 |
observation: RedTeamObservation
|
| 64 |
reward: float
|
| 65 |
|
| 66 |
+
|
| 67 |
class ResetResponse(BaseModel):
|
| 68 |
observation: RedTeamObservation
|
| 69 |
episode_id: str
|