cc.ja.300.binのダウンロードタイミングをDockerビルド時に変更

api/main.py

@@ -11,7 +11,6 @@ from pydantic import BaseModel
 from dotenv import load_dotenv
 
 from utils.logger import setup_logger
-from utils.json import get_file_path_from_config
 from api.search.engine import SearchEngine
 import schemas.projects as schema_projects
 
@@ -36,39 +35,11 @@ async def get_api_key(key: str = Security(api_key_header)):
 engine = SearchEngine()
 
 
-def download_embeddings() -> None:
-    """Hugging Face Datasets から Embeddings をダウンロードする"""
-    from huggingface_hub import hf_hub_download
-
-    # フォルダがなかったら新規作成
-    target_dir = "resources/embeddings"
-    os.makedirs(target_dir, exist_ok=True)
-
-    # ダウンロード
-    path = hf_hub_download(
-        repo_id=os.getenv("HF_EMBEDDINGS_REPO_ID"),
-        repo_type="dataset",
-        filename="cc.ja.300.bin",
-        local_dir=target_dir,
-        local_dir_use_symlinks=False,
-        token=os.getenv("HF_TOKEN"),
-    )
-
-    log.info(f"Embeddings downloaded: {path}")
-    return
-
-
 @asynccontextmanager
 async def lifespan(app: FastAPI):
     """アプリケーションの起動時と終了時に実行されるコード"""
     # --- 起動時処理 ---
     log.info("Initializing search engine and loading assets...")
-    load_dotenv()
-    if os.path.exists(get_file_path_from_config("embeddings.fasttext_bin")):
-        log.info("FastText embeddings found locally.")
-    else:
-        log.info("FastText embeddings not found locally. Downloading...")
-        download_embeddings()
     engine.initialize()
     log.info("Initialization complete.")
     yield

scripts/0_download_data.py

@@ -0,0 +1,38 @@
+import os
+
+from utils.logger import setup_logger
+from utils.json import get_file_path_from_config
+
+log = setup_logger(__name__)
+
+
+def download_embeddings() -> None:
+    """Hugging Face Datasets から Embeddings をダウンロードする"""
+    from huggingface_hub import hf_hub_download
+
+    # フォルダがなかったら新規作成
+    target_dir = "resources/embeddings"
+    os.makedirs(target_dir, exist_ok=True)
+
+    log.info("Downloading embeddings from Hugging Face Datasets...")
+
+    # ダウンロード
+    path = hf_hub_download(
+        repo_id=os.getenv("HF_EMBEDDINGS_REPO_ID"),
+        repo_type="dataset",
+        filename="cc.ja.300.bin",
+        local_dir=target_dir,
+        local_dir_use_symlinks=False,
+        token=os.getenv("HF_TOKEN"),
+    )
+
+    log.info(f"Embeddings downloaded: {path}")
+    return
+
+
+if __name__ == "__main__":
+    fast_text_bin_path = get_file_path_from_config("embeddings.fasttext_bin")
+    if os.path.exists(fast_text_bin_path):
+        log.info("Embeddings already exist. Skipping download.")
+    else:
+        download_embeddings()

scripts/build_all.py

@@ -24,6 +24,9 @@ def run_step(cmd: list[str], allow_fail: bool = False):
 def main():
     files = field_getter("config/files.json")
 
+    # Step 0: Download embeddings
+    run_step([sys.executable, "scripts/0_download_data.py"], allow_fail=True)
+
     # Step 1: Sudachi user dict (optional but recommended before tokenization)
     run_step([sys.executable, "scripts/1_build_dict.py"], allow_fail=True)
 
@@ -42,7 +45,9 @@ def main():
     # Step 6: embeddings (.vec がある場合のみ)
     vec_path = files("embeddings.fasttext_vec")
     if os.path.exists(vec_path):
-        run_step([sys.executable, "scripts/6_build_word_embeddings.py"])  # needs tf_token, bm25_meta
+        run_step(
+            [sys.executable, "scripts/6_build_word_embeddings.py"]
+        )  # needs tf_token, bm25_meta
     else:
         log.info(".vec が見つからないため Step 6 をスキップします: %s", vec_path)