synth-ai 0.2.4.dev4__py3-none-any.whl → 0.2.4.dev6__py3-none-any.whl

synth_ai/environments/examples/tictactoe/taskset.py

@@ -0,0 +1,214 @@
+from __future__ import annotations
+
+import random
+from dataclasses import dataclass
+from typing import List
+import numpy as np
+
+from uuid import uuid4
+from synth_ai.environments.tasks.core import (
+    TaskInstance,
+    TaskInstanceMetadata,
+    TaskInstanceSet,
+    Impetus,
+    Intent,
+    SplitInfo,
+)
+
+from .engine import COORD_TO_IDX, WIN_PATTERNS, PLAYER_MARKS
+
+
+@dataclass
+class TicTacToeTaskInstanceMetadata(TaskInstanceMetadata):
+    starting_player: str  # "X" or "O"
+    opening_moves: List[str]  # Pre-made moves to create position
+    optimal_outcome: str  # "win", "draw", "loss" for starting player
+    position_complexity: int  # Number of pre-moves made
+    shortest_win_length: int  # Min moves to force win/draw
+
+
+@dataclass
+class TicTacToeTaskInstance(TaskInstance):
+    async def serialize(self) -> dict:
+        return {
+            "id": str(self.id),
+            "impetus": {"instructions": self.impetus.instructions},
+            "intent": {
+                "rubric": self.intent.rubric,
+                "gold_trajectories": self.intent.gold_trajectories,
+                "gold_state_diff": self.intent.gold_state_diff,
+            },
+            "metadata": {
+                "starting_player": self.metadata.starting_player,
+                "opening_moves": self.metadata.opening_moves,
+                "optimal_outcome": self.metadata.optimal_outcome,
+                "position_complexity": self.metadata.position_complexity,
+                "shortest_win_length": self.metadata.shortest_win_length,
+            },
+            "is_reproducible": self.is_reproducible,
+            "initial_engine_snapshot": self.initial_engine_snapshot,
+        }
+
+    @classmethod
+    async def deserialize(cls, data: dict) -> "TicTacToeTaskInstance":
+        from uuid import UUID
+
+        metadata = TicTacToeTaskInstanceMetadata(
+            starting_player=data["metadata"]["starting_player"],
+            opening_moves=data["metadata"]["opening_moves"],
+            optimal_outcome=data["metadata"]["optimal_outcome"],
+            position_complexity=data["metadata"]["position_complexity"],
+            shortest_win_length=data["metadata"]["shortest_win_length"],
+        )
+
+        return cls(
+            id=UUID(data["id"]),
+            impetus=Impetus(instructions=data["impetus"]["instructions"]),
+            intent=Intent(
+                rubric=data["intent"]["rubric"],
+                gold_trajectories=data["intent"]["gold_trajectories"],
+                gold_state_diff=data["intent"]["gold_state_diff"],
+            ),
+            metadata=metadata,
+            is_reproducible=data["is_reproducible"],
+            initial_engine_snapshot=data["initial_engine_snapshot"],
+        )
+
+
+def _evaluate_position(board: np.ndarray, player: int) -> str:
+    """Simple evaluation of position outcome with perfect play"""
+    # Check for immediate win
+    for pattern in WIN_PATTERNS:
+        values = [board[i] for i in pattern]
+        if values.count(player) == 3:
+            return "win"
+        if values.count(3 - player) == 3:
+            return "loss"
+
+    # Check if board is full
+    if np.all(board != 0):
+        return "draw"
+
+    # For simplicity, assume draw for non-terminal positions
+    # In a real implementation, this would use minimax
+    return "draw"
+
+
+def _count_shortest_win(board: np.ndarray, player: int) -> int:
+    """Count minimum moves to force a win/draw"""
+    # Simplified: return remaining empty cells
+    empty_cells = sum(1 for i in range(9) if board[i] == 0)
+    return max(1, empty_cells // 2)
+
+
+async def create_tictactoe_taskset() -> TaskInstanceSet:
+    """Generate diverse TicTacToe starting positions"""
+    instances = []
+
+    # Configuration for different position types
+    POSITION_CONFIGS = {
+        "opening": {"pre_moves": 0, "count": 10},  # Fresh games
+        "early": {"pre_moves": 1, "count": 15},  # After 1 move
+        "mid": {"pre_moves": 2, "count": 15},  # After 2 moves
+        "complex": {"pre_moves": 3, "count": 10},  # After 3 moves
+    }
+
+    all_coords = list(COORD_TO_IDX.keys())
+
+    for config_name, config in POSITION_CONFIGS.items():
+        for i in range(config["count"]):
+            # Generate random opening moves
+            opening_moves = []
+            board = np.zeros(9, dtype=int)
+            current_player = "X"
+
+            # Make pre-moves
+            available_coords = all_coords.copy()
+            for move_idx in range(config["pre_moves"]):
+                if not available_coords:
+                    break
+
+                # Random move
+                move = random.choice(available_coords)
+                opening_moves.append(move)
+                available_coords.remove(move)
+
+                # Update board
+                board[COORD_TO_IDX[move]] = PLAYER_MARKS[current_player]
+                current_player = "O" if current_player == "X" else "X"
+
+            # Evaluate position
+            starting_player = current_player
+            optimal_outcome = _evaluate_position(board, PLAYER_MARKS[starting_player])
+            shortest_win = _count_shortest_win(board, PLAYER_MARKS[starting_player])
+
+            # Create metadata
+            metadata = TicTacToeTaskInstanceMetadata(
+                starting_player=starting_player,
+                opening_moves=opening_moves,
+                optimal_outcome=optimal_outcome,
+                position_complexity=config["pre_moves"],
+                shortest_win_length=shortest_win,
+            )
+
+            # Create instance
+            impetus = Impetus(
+                instructions=(
+                    f"You are playing TicTacToe as {starting_player}. "
+                    + "The game is played on a 3x3 grid with cells labeled A1-A3, B1-B3, C1-C3. "
+                    + (
+                        f"The game has already had {len(opening_moves)} moves."
+                        if opening_moves
+                        else "This is a fresh game."
+                    )
+                    + f" You must place your mark ({starting_player}) in an empty cell. "
+                    + "Win by getting three of your marks in a row (horizontally, vertically, or diagonally)."
+                )
+            )
+
+            intent = Intent(
+                rubric={"goal": f"Win the game as {starting_player}, or at least force a draw"},
+                gold_trajectories=None,
+                gold_state_diff={"optimal_outcome": optimal_outcome},
+            )
+
+            instance = TicTacToeTaskInstance(
+                id=uuid4(),
+                impetus=impetus,
+                intent=intent,
+                metadata=metadata,
+                is_reproducible=True,
+                initial_engine_snapshot=None,
+            )
+
+            instances.append(instance)
+
+    # Shuffle instances
+    random.shuffle(instances)
+
+    # Define splits based on complexity
+    val_ids = {inst.id for inst in instances if inst.metadata.position_complexity == 1}
+    test_ids = {inst.id for inst in instances if inst.metadata.position_complexity >= 2}
+
+    # If not enough instances for splits, use simple division
+    if len(val_ids) == 0 or len(test_ids) == 0:
+        total = len(instances)
+        val_end = int(total * 0.15)
+        test_end = int(total * 0.30)
+        val_ids = {instances[i].id for i in range(val_end)}
+        test_ids = {instances[i].id for i in range(val_end, test_end)}
+
+    split_info = SplitInfo(
+        val_instance_ids=val_ids, test_instance_ids=test_ids, _is_split_defined=True
+    )
+
+    return TaskInstanceSet(
+        name="TicTacToe Procedural TaskSet",
+        description="Procedurally generated TicTacToe tasks with varying starting positions.",
+        instances=instances,
+        split_info=split_info,
+    )
+
+
+# Make taskset available as module attribute
+taskset = create_tictactoe_taskset

synth_ai/environments/examples/verilog/__init__.py

@@ -0,0 +1,10 @@
+from .engine import VerilogEngine
+from .environment import VerilogEnvironment
+from .taskset import VerilogTaskInstance, create_verilog_taskset
+
+__all__ = [
+    "VerilogEngine",
+    "VerilogEnvironment",
+    "VerilogTaskInstance",
+    "create_verilog_taskset",
+]

synth_ai/environments/examples/verilog/engine.py

@@ -0,0 +1,328 @@
+from __future__ import annotations
+import shutil
+import subprocess
+from pathlib import Path
+from typing import Dict, Any, Tuple, Optional
+from dataclasses import dataclass
+
+from synth_ai.environments.stateful.engine import StatefulEngine, StatefulEngineSnapshot
+from synth_ai.environments.tasks.core import TaskInstance
+from synth_ai.environments.environment.rewards.core import RewardStack, RewardComponent
+
+
+@dataclass
+class VerilogEngineSnapshot(StatefulEngineSnapshot):
+    task_instance_dict: Dict
+    engine_snapshot: Dict
+
+    def model_dump(self) -> Dict:
+        """Convert dataclass to dictionary for compatibility with Pydantic models."""
+        return {
+            "task_instance_dict": self.task_instance_dict,
+            "engine_snapshot": self.engine_snapshot,
+        }
+
+
+@dataclass
+class VerilogPublicState:
+    files: Dict[str, str]
+    build_dir: str
+    task_completed: bool = False
+    last_compile_output: Optional[str] = None
+    last_simulate_output: Optional[str] = None
+
+
+@dataclass
+class VerilogPrivateState:
+    reward_last: float
+    total_reward: float
+    terminated: bool
+    truncated: bool
+
+
+class VerilogCompileSuccessComponent(RewardComponent):
+    async def score(self, state: VerilogPublicState, action: Any) -> float:
+        if hasattr(action, "get") and action.get("type") == "compile":
+            # Check if compilation was successful (returncode 0)
+            if action.get("returncode") == 0:
+                return 0.1
+        return 0.0
+
+
+class VerilogSimulationPassComponent(RewardComponent):
+    async def score(self, state: VerilogPublicState, action: Any) -> float:
+        if hasattr(action, "get") and action.get("type") == "simulate":
+            # Check if simulation passed
+            if action.get("passed", False):
+                return 1.0
+        return 0.0
+
+
+class VerilogStepPenaltyComponent(RewardComponent):
+    def __init__(self, penalty: float = -0.01):
+        self.penalty = penalty
+
+    async def score(self, state: Any, action: Any) -> float:
+        return self.penalty
+
+
+class VerilogEngine(StatefulEngine):
+    """
+    Stateful Verilog evaluation engine with persistent artifact snapshots.
+    """
+
+    def __init__(self, task_instance: TaskInstance):
+        self.task_instance = task_instance
+        self._total_reward = 0.0
+        self._current_action_for_reward: Optional[Dict[str, Any]] = None
+
+        self.reward_stack = RewardStack(
+            components=[
+                VerilogCompileSuccessComponent(),
+                VerilogSimulationPassComponent(),
+                VerilogStepPenaltyComponent(penalty=-0.01),
+            ]
+        )
+
+        # Initialize paths - will be set properly in _reset_engine
+        self.snapshot_dir: Optional[Path] = None
+        self.build_dir: Optional[Path] = None
+
+        # Track last compile/simulate outputs
+        self._last_compile_output: Optional[str] = None
+        self._last_simulate_output: Optional[str] = None
+
+    async def _reset_engine(
+        self, *, seed: Optional[int] = None
+    ) -> Tuple[VerilogPrivateState, VerilogPublicState]:
+        """Initialize the Verilog environment with task files."""
+        self._total_reward = 0.0
+        self._current_action_for_reward = None
+        self._last_compile_output = None
+        self._last_simulate_output = None
+
+        # Initialize snapshot from task instance
+        self._init_snapshot()
+
+        priv = VerilogPrivateState(
+            reward_last=0.0, total_reward=0.0, terminated=False, truncated=False
+        )
+
+        pub = VerilogPublicState(
+            files=self._get_file_contents(),
+            build_dir=str(self.build_dir),
+            task_completed=False,
+        )
+
+        return priv, pub
+
+    async def _step_engine(
+        self, action_result: Dict[str, Any]
+    ) -> Tuple[VerilogPrivateState, VerilogPublicState]:
+        """Process an action result and update engine state."""
+        self._current_action_for_reward = action_result
+
+        # Update last outputs if this is a compile or simulate action
+        if action_result.get("type") == "compile":
+            stdout = action_result.get("stdout", "")
+            stderr = action_result.get("stderr", "")
+            # Combine stdout and stderr for compile output, stderr has the error info
+            self._last_compile_output = stderr if stderr else stdout
+        elif action_result.get("type") == "simulate":
+            self._last_simulate_output = action_result.get("stdout")
+
+        # Calculate reward using RewardStack
+        current_pub_state = VerilogPublicState(
+            files=self._get_file_contents(),
+            build_dir=str(self.build_dir),
+            task_completed=action_result.get("passed", False),
+        )
+
+        reward_from_stack = await self.reward_stack.step_reward(
+            state=current_pub_state, action=self._current_action_for_reward
+        )
+        self._current_action_for_reward = None
+
+        self._total_reward += reward_from_stack
+
+        # Check termination conditions
+        terminated = action_result.get("passed", False) or action_result.get("submitted", False)
+
+        priv = VerilogPrivateState(
+            reward_last=reward_from_stack,
+            total_reward=self._total_reward,
+            terminated=terminated,
+            truncated=False,
+        )
+
+        pub = VerilogPublicState(
+            files=self._get_file_contents(),
+            build_dir=str(self.build_dir),
+            task_completed=action_result.get("passed", False),
+            last_compile_output=self._last_compile_output,
+            last_simulate_output=self._last_simulate_output,
+        )
+
+        return priv, pub
+
+    def _init_snapshot(self) -> None:
+        """Initialize snapshot directory from task instance data."""
+        if not hasattr(self.task_instance, "snapshot_dir"):
+            raise ValueError("Task instance must have a snapshot_dir attribute")
+
+        self.snapshot_dir = Path(self.task_instance.snapshot_dir)
+
+        if self.snapshot_dir.exists() and any(self.snapshot_dir.iterdir()):
+            # Already initialized
+            self.build_dir = self.snapshot_dir / "build"
+            self.build_dir.mkdir(exist_ok=True)
+            return
+
+        # Copy pristine files from task data
+        pristine_dir = getattr(self.task_instance, "pristine_dir", None)
+        if pristine_dir and Path(pristine_dir).exists():
+            shutil.copytree(pristine_dir, self.snapshot_dir, dirs_exist_ok=True)
+        else:
+            # Create basic structure if no pristine dir
+            self.snapshot_dir.mkdir(parents=True, exist_ok=True)
+
+        self.build_dir = self.snapshot_dir / "build"
+        self.build_dir.mkdir(exist_ok=True)
+
+    def _get_file_contents(self) -> Dict[str, str]:
+        """Get contents of all Verilog files in the snapshot directory."""
+        if not self.snapshot_dir:
+            return {}
+
+        files = {}
+        for p in self.snapshot_dir.rglob("*.v"):
+            try:
+                relative_path = p.relative_to(self.snapshot_dir)
+                files[str(relative_path)] = p.read_text()
+            except Exception:
+                continue
+        return files
+
+    async def write_file(self, path: str, content: str) -> Dict[str, Any]:
+        """Write content to a file in the snapshot directory."""
+        if not self.snapshot_dir:
+            return {"ok": False, "error": "Snapshot directory not initialized"}
+
+        file_path = self.snapshot_dir / path
+        file_path.parent.mkdir(parents=True, exist_ok=True)
+        file_path.write_text(content)
+        return {"ok": True, "type": "write_file"}
+
+    async def compile(
+        self, sources: Optional[list] = None, testbench: Optional[str] = None
+    ) -> Dict[str, Any]:
+        """Compile Verilog sources with iverilog."""
+        if not self.snapshot_dir or not self.build_dir:
+            return {"ok": False, "error": "Directories not initialized"}
+
+        # Default to all .v files if no sources specified
+        if sources is None:
+            sources = [str(p.relative_to(self.snapshot_dir)) for p in self.snapshot_dir.glob("*.v")]
+
+        src_paths = [self.snapshot_dir / src for src in sources]
+
+        # Add testbench if specified
+        if testbench:
+            tb_path = self.snapshot_dir / testbench
+            if tb_path.exists():
+                src_paths.append(tb_path)
+
+        binary = self.build_dir / "a.out"
+        cmd = ["iverilog", "-g2012", "-o", str(binary)] + [str(p) for p in src_paths]
+
+        try:
+            proc = subprocess.run(cmd, capture_output=True, text=True, timeout=30)
+            return {
+                "ok": proc.returncode == 0,
+                "type": "compile",
+                "stdout": proc.stdout,
+                "stderr": proc.stderr,
+                "returncode": proc.returncode,
+                "binary": str(binary) if proc.returncode == 0 else None,
+            }
+        except subprocess.TimeoutExpired:
+            return {"ok": False, "error": "Compilation timeout", "type": "compile"}
+        except Exception as e:
+            return {"ok": False, "error": str(e), "type": "compile"}
+
+    async def simulate(self, binary: Optional[str] = None) -> Dict[str, Any]:
+        """Run vvp on compiled binary."""
+        if not self.build_dir:
+            return {"ok": False, "error": "Build directory not initialized"}
+
+        bin_path = binary if binary else str(self.build_dir / "a.out")
+
+        try:
+            proc = subprocess.run(["vvp", bin_path], capture_output=True, text=True, timeout=30)
+
+            # Check for various success indicators
+            stdout = proc.stdout
+            passed = (
+                "ALL_TESTS_PASSED" in stdout
+                or ("Mismatches: 0 " in stdout and "samples" in stdout)
+                or ("no mismatches" in stdout.lower() and "errors" not in stdout.lower())
+            )
+
+            return {
+                "ok": True,
+                "type": "simulate",
+                "stdout": proc.stdout,
+                "stderr": proc.stderr,
+                "returncode": proc.returncode,
+                "passed": passed,
+            }
+        except subprocess.TimeoutExpired:
+            return {"ok": False, "error": "Simulation timeout", "type": "simulate"}
+        except Exception as e:
+            return {"ok": False, "error": str(e), "type": "simulate"}
+
+    async def submit(self) -> Dict[str, Any]:
+        """Submit solution for grading."""
+        # For now, simple check based on last simulation
+        # In a full implementation, this would call the task's verify method
+        return {
+            "ok": True,
+            "type": "submit",
+            "passed": True,  # Placeholder
+            "detail": "Submission processed",
+            "submitted": True,
+        }
+
+    async def _serialize_engine(self) -> VerilogEngineSnapshot:
+        """Serialize engine state to a snapshot."""
+        engine_data = {
+            "total_reward": self._total_reward,
+            "snapshot_dir": str(self.snapshot_dir) if self.snapshot_dir else None,
+            "build_dir": str(self.build_dir) if self.build_dir else None,
+        }
+
+        task_instance_dict = await self.task_instance.serialize()
+
+        return VerilogEngineSnapshot(
+            task_instance_dict=task_instance_dict, engine_snapshot=engine_data
+        )
+
+    @classmethod
+    async def _deserialize_engine(cls, snapshot: VerilogEngineSnapshot) -> "VerilogEngine":
+        """Deserialize engine from snapshot."""
+        # This would need proper task instance deserialization
+        # For now, create a minimal implementation
+        from synth_ai.environments.examples.verilog.taskset import VerilogTaskInstance
+
+        task_instance = await VerilogTaskInstance.deserialize(snapshot.task_instance_dict)
+        engine = cls(task_instance)
+
+        engine_data = snapshot.engine_snapshot
+        engine._total_reward = engine_data.get("total_reward", 0.0)
+
+        if engine_data.get("snapshot_dir"):
+            engine.snapshot_dir = Path(engine_data["snapshot_dir"])
+        if engine_data.get("build_dir"):
+            engine.build_dir = Path(engine_data["build_dir"])
+
+        return engine