Spaces:
Sleeping
Sleeping
File size: 11,361 Bytes
b196357 86c3e08 b196357 cf05092 b196357 86c3e08 b196357 bc02c39 b196357 86c3e08 b196357 bc02c39 b196357 bc02c39 b196357 bc02c39 b196357 bc02c39 b196357 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 | from __future__ import annotations
from dataclasses import dataclass, field
from pathlib import Path
from uuid import uuid4
from pydantic import BaseModel, ConfigDict
from sqlmodel import Session, select
from db.schema import ModuleNode
from db.seed import seed_project
from db.store import Store
from env.action import ActionType, ReviewAction
from env.observation import CodeObservation
from env.observation_builder import ObservationBuilder
from env.runtime_config import RuntimeConfig, load_runtime_config
from env.state import EpisodeState, GraphState, ModuleReviewState
from graph.graph_manager import GraphManager
from graders.base_grader import EpisodeState as GraderEpisodeState
from graders.base_grader import BaseGrader
from graders.easy_grader import EasyGrader
from graders.hard_grader import HardGrader
from graders.medium_grader import MediumGrader
from llm.lora_adapter import LoRATrajectoryLogger
from tasks.task_registry import TaskSpec, get_task, list_tasks, resolve_task_modules
class StepResult(BaseModel):
model_config = ConfigDict(strict=True, extra="forbid")
observation: CodeObservation
reward: float
done: bool
feedback: str
episode_id: str
@dataclass
class _EpisodeRuntime:
episode_id: str
task: TaskSpec
modules: list[str]
module_index: int = 0
step_count: int = 0
cumulative_reward: float = 0.0
done: bool = False
module_steps: dict[str, int] = field(default_factory=dict)
module_rewards: dict[str, float] = field(default_factory=dict)
module_last_action: dict[str, str] = field(default_factory=dict)
module_last_reward: dict[str, float] = field(default_factory=dict)
grader_states: dict[str, GraderEpisodeState] = field(default_factory=dict)
@property
def current_module(self) -> str | None:
if self.done:
return None
if self.module_index >= len(self.modules):
return None
return self.modules[self.module_index]
@property
def modules_remaining(self) -> int:
return max(len(self.modules) - self.module_index, 0)
class CodeReviewEnv:
"""OpenEnv-style environment runtime backed by persistent SQLite state."""
SUPPORTS_CONCURRENT_SESSIONS = False
def __init__(
self,
source_root: str | Path = "sample_project",
db_path: str | Path | None = None,
runtime_config: RuntimeConfig | None = None,
) -> None:
self.source_root = str(Path(source_root).resolve())
self.db_path = str(db_path) if db_path is not None else None
self.config = runtime_config or load_runtime_config()
self.store = Store(source_root=self.source_root, db_path=self.db_path)
self.graph_manager = GraphManager(source_root=self.source_root, db_path=self.db_path)
self.observation_builder = ObservationBuilder(source_root=self.source_root, db_path=self.db_path)
self.lora_logger = LoRATrajectoryLogger()
self._runtime: _EpisodeRuntime | None = None
self._grader: BaseGrader | None = None
def available_tasks(self) -> list[TaskSpec]:
return list_tasks()
def reset(
self,
task_id: str = "style_review",
module_override: list[str] | None = None,
seed: int | None = None,
episode_id: str | None = None,
) -> CodeObservation:
del seed
if not self.store.has_nodes():
seed_project(target_dir=Path(self.source_root), db_path=self.db_path, force=False)
self.graph_manager.invalidate_cache()
task = get_task(task_id)
modules = resolve_task_modules(task, self.graph_manager, module_override=module_override)
if not modules:
raise ValueError(f"Task {task_id} has no resolvable modules in current graph")
runtime = _EpisodeRuntime(
episode_id=episode_id or str(uuid4()),
task=task,
modules=modules,
module_steps={module_id: 0 for module_id in modules},
module_rewards={module_id: 0.0 for module_id in modules},
grader_states={module_id: GraderEpisodeState() for module_id in modules},
)
self._runtime = runtime
self._grader = self._create_grader(task.grader)
for module_id in modules:
self.store.create_episode_record(runtime.episode_id, task.task_id, module_id)
return self.observation_builder.build(
module_id=modules[0],
task_description=task.description,
)
def step(self, action: ReviewAction) -> StepResult:
if self._runtime is None or self._grader is None:
raise RuntimeError("Environment not initialized. Call reset() before step().")
runtime = self._runtime
module_id = runtime.current_module
if module_id is None:
runtime.done = True
observation = self._build_terminal_observation(runtime)
return StepResult(
observation=observation,
reward=0.0,
done=True,
feedback="Episode already complete",
episode_id=runtime.episode_id,
)
findings = self._grader._sorted_findings(module_id)
grader_state = runtime.grader_states[module_id]
reward = self._grader.grade_action(
module_id=module_id,
action=action,
findings=findings,
state=grader_state,
)
grader_state.seen_actions.append(action)
step_number = runtime.step_count + 1
self._grader._persist_step(
module_id=module_id,
task_id=runtime.task.task_id,
episode_id=runtime.episode_id,
step_number=step_number,
action=action,
reward=reward,
)
runtime.step_count = step_number
runtime.cumulative_reward += reward.raw_value
runtime.module_steps[module_id] = runtime.module_steps.get(module_id, 0) + 1
runtime.module_rewards[module_id] = runtime.module_rewards.get(module_id, 0.0) + reward.raw_value
runtime.module_last_action[module_id] = action.action_type.value
runtime.module_last_reward[module_id] = reward.raw_value
self.store.update_episode_record(
episode_id=runtime.episode_id,
module_id=module_id,
total_steps=runtime.module_steps[module_id],
cumulative_reward=runtime.module_rewards[module_id],
)
end_module = action.action_type in {ActionType.APPROVE, ActionType.REQUEST_CHANGES}
if end_module:
runtime.module_index += 1
if runtime.module_index >= len(runtime.modules):
runtime.done = True
if runtime.step_count >= max(runtime.task.max_steps, self.config.max_steps_per_episode):
runtime.done = True
next_module = runtime.current_module
if runtime.done or next_module is None:
observation = self._build_terminal_observation(runtime)
else:
context_request = action.context_request if action.action_type == ActionType.REQUEST_CONTEXT else None
if context_request is not None:
context_request = self.graph_manager.resolve_module_id(context_request)
observation = self.observation_builder.build(
module_id=next_module,
task_description=runtime.task.description,
context_request=context_request,
)
self.lora_logger.log(
source_root=self.source_root,
episode_id=runtime.episode_id,
module_id=module_id,
step_number=step_number,
action=action,
reward=reward.raw_value,
done=runtime.done,
task_id=runtime.task.task_id,
observation_summary=f"module={observation.module_id} actions={','.join(observation.available_actions[:6])}",
)
return StepResult(
observation=observation,
reward=reward.raw_value,
done=runtime.done,
feedback=reward.feedback,
episode_id=runtime.episode_id,
)
def reset_episode_annotations(self) -> int:
if self._runtime is None:
return 0
return self.store.clear_annotations_for_episode(self._runtime.episode_id)
def state(self) -> GraphState:
if self._runtime is None:
raise RuntimeError("Environment not initialized. Call reset() first.")
runtime = self._runtime
snapshot = self.store.get_full_graph()
annotations = self.store.get_review_annotations(episode_id=runtime.episode_id)
all_annotations = self.store.get_review_annotations()
last_by_module: dict[str, tuple[str, float]] = {}
for annotation in annotations:
last_by_module[annotation.module_id] = (annotation.action_type, annotation.reward_given)
modules: list[ModuleReviewState] = []
with Session(self.store.engine) as session:
node_rows = list(
session.exec(
select(ModuleNode).where(ModuleNode.source_root == self.store.config.source_root)
).all()
)
for node in sorted(node_rows, key=lambda item: item.module_id):
action_name, last_reward = last_by_module.get(node.module_id, (None, 0.0))
modules.append(
ModuleReviewState(
module_id=node.module_id,
module_summary=node.summary,
review_status=node.review_status.value,
latest_review_summary=node.review_summary,
issues_found=len(self.store.get_findings(node.module_id)),
last_action=action_name,
last_reward=last_reward,
updated_at=node.updated_at.isoformat(),
)
)
return GraphState(
episode=EpisodeState(
episode_id=runtime.episode_id,
task_id=runtime.task.task_id,
current_module_id=runtime.current_module,
modules_remaining=runtime.modules_remaining,
step_count=runtime.step_count,
cumulative_reward=runtime.cumulative_reward,
done=runtime.done,
status="complete" if runtime.done else "running",
),
modules=modules,
module_count=len(snapshot.nodes),
edge_count=len(snapshot.edges),
annotation_count=len(annotations),
total_annotation_count=len(all_annotations),
)
def _create_grader(self, grader_name: str) -> BaseGrader:
if grader_name == "easy":
return EasyGrader(self.store)
if grader_name == "medium":
return MediumGrader(self.store)
if grader_name == "hard":
return HardGrader(self.store, self.graph_manager)
raise ValueError(f"Unknown grader type: {grader_name}")
def _build_terminal_observation(self, runtime: _EpisodeRuntime) -> CodeObservation:
fallback = runtime.modules[-1]
return self.observation_builder.build(
module_id=fallback,
task_description=f"{runtime.task.description} (episode complete)",
)
|