Update app.py

app.py

@@ -7,13 +7,14 @@ import json
 import os
 import threading
 import html
+import shutil
 from collections import defaultdict
 from datetime import datetime
 from pathlib import Path
 from typing import Any, Dict, List, Optional, Tuple
 
 import gradio as gr
-from huggingface_hub import CommitScheduler, snapshot_download
+from huggingface_hub import CommitScheduler, HfApi, snapshot_download
 
 # 路径配置（按用户要求）
 # Spaces 推荐优先读取当前 Space 仓库内文件（app.py 同级）
@@ -35,6 +36,7 @@ VIDEO_DIR = INPUT_DIR / "video"
 
 Path(OUTPUT_DIR).mkdir(parents=True, exist_ok=True)
 scheduler: Optional[CommitScheduler] = None
+hf_api = HfApi()
 
 
 def _set_paths(input_dir: Path, output_dir: Path) -> None:
@@ -95,7 +97,8 @@ def _try_download_from_hub() -> bool:
     else:
         return False
 
-    hub_output = hub_root / "user_study_results"
+    # 结果统一写到 Space 仓库内目录，避免写到缓存目录后用户难以定位
+    hub_output = LOCAL_OUTPUT_DIR
     _set_paths(hub_input, hub_output)
     return True
 
@@ -391,8 +394,47 @@ def load_evaluated_method_story_pairs() -> set:
     return evaluated
 
 
+def sync_results_from_hub_to_local() -> None:
+    """
+    从远程结果仓库拉取最新结果到本地 OUTPUT_DIR。
+    仅用于“判定哪些样本已评估”，保证展示逻辑以远程为准。
+    """
+    if not RESULTS_REPO_ID:
+        return
+    sync_root = APP_DIR / ".hf_results_sync_cache"
+    try:
+        snapshot_download(
+            repo_id=RESULTS_REPO_ID,
+            repo_type="dataset",
+            local_dir=str(sync_root),
+            token=HF_TOKEN,
+            allow_patterns=["user_study_results/**"],
+        )
+    except Exception as e:
+        print(f"[SYNC] pull results repo failed: {e}")
+        return
+
+    remote_results_root = sync_root / "user_study_results"
+    if not remote_results_root.exists():
+        return
+
+    remote_raw = remote_results_root / "raw_results"
+    local_raw = OUTPUT_DIR / "raw_results"
+    if local_raw.exists():
+        shutil.rmtree(local_raw)
+    if remote_raw.exists():
+        shutil.copytree(remote_raw, local_raw)
+
+    remote_agg = remote_results_root / "method_aggregates.json"
+    local_agg = OUTPUT_DIR / "method_aggregates.json"
+    if remote_agg.exists():
+        local_agg.parent.mkdir(parents=True, exist_ok=True)
+        shutil.copy2(remote_agg, local_agg)
+
+
 def build_pending_samples() -> List[Dict[str, Any]]:
     """构建待评估样本池，并分配匿名ID。"""
+    sync_results_from_hub_to_local()
     all_samples = load_dataset_index()
     evaluated_pairs = load_evaluated_method_story_pairs()
     pending = [
@@ -502,6 +544,39 @@ def recompute_method_aggregates() -> Path:
     return out_path
 
 
+def push_result_files_to_hub(single_path: Path, agg_path: Path) -> Optional[str]:
+    """
+    提交后立即把结果文件上传到 RESULTS_REPO_ID，避免仅依赖定时 CommitScheduler。
+    返回 None 表示成功；返回字符串表示失败原因。
+    """
+    if not RESULTS_REPO_ID:
+        return "未配置 RESULTS_REPO_ID。"
+    if not HF_TOKEN:
+        return "未配置 HF_TOKEN，无法写入 Hugging Face 远程仓库。"
+
+    try:
+        single_rel = single_path.relative_to(OUTPUT_DIR).as_posix()
+        hf_api.upload_file(
+            path_or_fileobj=str(single_path),
+            path_in_repo=f"user_study_results/{single_rel}",
+            repo_id=RESULTS_REPO_ID,
+            repo_type="dataset",
+            token=HF_TOKEN,
+        )
+
+        agg_rel = agg_path.relative_to(OUTPUT_DIR).as_posix()
+        hf_api.upload_file(
+            path_or_fileobj=str(agg_path),
+            path_in_repo=f"user_study_results/{agg_rel}",
+            repo_id=RESULTS_REPO_ID,
+            repo_type="dataset",
+            token=HF_TOKEN,
+        )
+        return None
+    except Exception as e:
+        return str(e)
+
+
 def build_sample_brief_html(sample: Dict[str, Any], index: int, total: int) -> str:
     story = sample.get("story_text") or "(未找到对应 story 文本，请检查 clip_movie_story 下是否有同名 txt)"
     safe_story = html.escape(story)
@@ -620,6 +695,7 @@ def create_app():
             evaluator_id = (evaluator_id or "anonymous").strip() or "anonymous"
 
             # 防重复：方法-故事只允许评估一次
+            sync_results_from_hub_to_local()
             evaluated_pairs = load_evaluated_method_story_pairs()
             if (sample["method"], sample["story_name"]) in evaluated_pairs:
                 return "⚠️ 该方法-故事已经被评估过一次，请选择其他匿名样本。"
@@ -636,6 +712,10 @@ def create_app():
             with SAVE_LOCK:
                 single_path = save_single_result(sample, evaluator_id, scores, reasons, summary or "")
                 agg_path = recompute_method_aggregates()
+                push_err = push_result_files_to_hub(single_path, agg_path)
+
+            if push_err:
+                return f"❌ 结果已本地保存，但写入远程 `{RESULTS_REPO_ID}` 失败：{push_err}"
 
             _ = (single_path, agg_path)
             return ""