Upload models.py with huggingface_hub

models.py

@@ -0,0 +1,73 @@
+"""
+Pydantic models for OpenEnv Tetris environment.
+Follows OpenEnv 0.2 spec: Action, Observation, StepResult.
+"""
+
+from pydantic import BaseModel, Field
+from typing import Optional
+from enum import Enum
+
+
+class ActionType(str, Enum):
+    LEFT = "left"
+    RIGHT = "right"
+    ROTATE_CW = "rotate_cw"
+    ROTATE_CCW = "rotate_ccw"
+    DROP = "drop"
+    DOWN = "down"
+    NOOP = "noop"
+
+
+class TetrisAction(BaseModel):
+    action: ActionType = Field(
+        ...,
+        description="Action to perform. One of: left, right, rotate_cw, rotate_ccw, drop, down, noop"
+    )
+
+
+class TetrisObservation(BaseModel):
+    board: str = Field(..., description="Text representation of the board (10x20 grid)")
+    current_piece: str = Field(..., description="Name of current piece (I, O, T, S, Z, L, J)")
+    current_piece_shape: str = Field(..., description="Text shape of current piece")
+    next_piece: str = Field(..., description="Name of next piece")
+    next_piece_shape: str = Field(..., description="Text shape of next piece")
+    piece_x: int = Field(..., description="Current piece X position")
+    piece_y: int = Field(..., description="Current piece Y position")
+    score: int = Field(..., description="Current score")
+    total_lines: int = Field(..., description="Total lines cleared")
+    steps: int = Field(..., description="Number of steps taken")
+    max_height: int = Field(..., description="Height of tallest column")
+    holes: int = Field(..., description="Number of holes in the board")
+
+
+class StepResult(BaseModel):
+    observation: TetrisObservation
+    reward: float = Field(..., description="Reward for this step")
+    done: bool = Field(..., description="Whether the episode is over")
+
+
+class ResetResult(BaseModel):
+    observation: TetrisObservation
+
+
+class ResetRequest(BaseModel):
+    seed: Optional[int] = Field(None, description="Optional random seed for reproducibility")
+
+
+class EnvInfo(BaseModel):
+    name: str = "tetris-env"
+    description: str = "Tetris environment for LLM agent training with combo scoring"
+    version: str = "0.1.0"
+    action_space: list[str] = ["left", "right", "rotate_cw", "rotate_ccw", "drop", "down", "noop"]
+    observation_format: str = "text"
+    board_size: str = "10x20"
+    reward_structure: dict = {
+        "1_line": 100,
+        "2_lines": 300,
+        "3_lines": 700,
+        "4_lines_tetris": 1500,
+        "step_penalty": -1,
+        "height_penalty": "-2 * max_height",
+        "hole_penalty": "-5 * holes",
+        "game_over": -500,
+    }