Upload folder using huggingface_hub

server/task_bank.py

@@ -3,6 +3,8 @@
 from __future__ import annotations
 
 from dataclasses import dataclass
+import json
+import os
 from typing import Dict, List, Tuple
 
 try:
@@ -85,6 +87,7 @@ TASKS: Dict[str, CodeReviewTask] = {
 
 
 EPISODE_SCORES: Dict[tuple[str, str], float] = {}
+SCORE_FILE = os.getenv("CODING_ENV_SCORE_FILE", "/tmp/coding_env_episode_scores.json")
 
 
 def list_tasks() -> List[Dict[str, str]]:
@@ -149,9 +152,48 @@ def grade_action(action: CodeAction, task: CodeReviewTask) -> tuple[float, str]:
 
 def record_episode_score(task_id: str, episode_id: str, score: float) -> None:
     """Persist normalized score for grader endpoint."""
-    EPISODE_SCORES[(task_id, episode_id)] = max(0.0, min(1.0, float(score)))
+    normalized = max(0.0, min(1.0, float(score)))
+    EPISODE_SCORES[(task_id, episode_id)] = normalized
+    _persist_score(task_id, episode_id, normalized)
 
 
 def get_episode_score(task_id: str, episode_id: str) -> float:
     """Read score for task/episode pair."""
-    return EPISODE_SCORES.get((task_id, episode_id), 0.0)
+    in_memory = EPISODE_SCORES.get((task_id, episode_id))
+    if in_memory is not None:
+        return in_memory
+    return _load_persisted_score(task_id, episode_id)
+
+
+def _persist_score(task_id: str, episode_id: str, score: float) -> None:
+    key = f"{task_id}::{episode_id}"
+    payload: Dict[str, float] = {}
+    if os.path.exists(SCORE_FILE):
+        try:
+            with open(SCORE_FILE, "r", encoding="utf-8") as f:
+                loaded = json.load(f)
+            if isinstance(loaded, dict):
+                payload = {
+                    str(k): float(v)
+                    for k, v in loaded.items()
+                    if isinstance(v, (int, float))
+                }
+        except Exception:
+            payload = {}
+
+    payload[key] = float(score)
+    with open(SCORE_FILE, "w", encoding="utf-8") as f:
+        json.dump(payload, f)
+
+
+def _load_persisted_score(task_id: str, episode_id: str) -> float:
+    if not os.path.exists(SCORE_FILE):
+        return 0.0
+    try:
+        with open(SCORE_FILE, "r", encoding="utf-8") as f:
+            loaded = json.load(f)
+        key = f"{task_id}::{episode_id}"
+        value = loaded.get(key, 0.0) if isinstance(loaded, dict) else 0.0
+        return max(0.0, min(1.0, float(value)))
+    except Exception:
+        return 0.0