untracked files on main: dceb101 feat(#63): IMP-34 R1 donor capacity measured bound (u1+u2)

2026-05-21 22:07:41 +09:00
commit 8f085a28d3
3220 changed files with 985495 additions and 0 deletions
@@ -0,0 +1,76 @@
+"""Phase 2: Semantic 매칭용 모델 다운로드"""
+import sys
+from pathlib import Path
+import urllib.request
+import zipfile
+
+MODELS_DIR = Path(r"d:\ad-hoc\kei\design_agent\models")
+MODELS_DIR.mkdir(exist_ok=True)
+
+
+def download_ko_sroberta():
+    """ko-sroberta-multitask (Sentence-BERT, 420MB)
+    sentence-transformers 라이브러리가 자동 캐시에 다운로드"""
+    print("\n=== [1/3] ko-sroberta-multitask 다운로드 ===")
+    from sentence_transformers import SentenceTransformer
+    model = SentenceTransformer('jhgan/ko-sroberta-multitask',
+                                 cache_folder=str(MODELS_DIR / "sentence-transformers"))
+    print(f"✓ 완료: 차원 {model.get_sentence_embedding_dimension()}")
+    return model
+
+
+def download_korean_fasttext():
+    """Korean FastText — Kyubyong 공개 모델 (100~200MB)
+    GitHub releases: https://github.com/Kyubyong/wordvectors"""
+    print("\n=== [2/3] Korean FastText 다운로드 ===")
+    # Kyubyong의 한국어 FastText 200차원 (약 400MB)
+    target = MODELS_DIR / "ko_fasttext"
+    target.mkdir(exist_ok=True)
+    model_file = target / "ko.bin"
+    if model_file.exists():
+        print(f"✓ 이미 존재: {model_file}")
+        return str(model_file)
+    # Facebook pre-trained FastText로 대체 - huggingface hub 사용
+    try:
+        from huggingface_hub import hf_hub_download
+        print("  HuggingFace에서 한국어 FastText 다운로드...")
+        path = hf_hub_download(
+            repo_id="facebook/fasttext-ko-vectors",
+            filename="model.bin",
+            cache_dir=str(MODELS_DIR / "fasttext"),
+        )
+        print(f"✓ 완료: {path}")
+        return path
+    except Exception as e:
+        print(f"  HuggingFace 실패: {e}")
+        # 대안: ko-sroberta만 써서 FastText 역할 커버
+        print("  → Korean FastText는 skip하고 ko-sroberta로 대체")
+        return None
+
+
+def download_korean_word2vec():
+    """Korean Word2Vec — gensim 데이터나 HF에서 받기"""
+    print("\n=== [3/3] Korean Word2Vec 다운로드 ===")
+    # 작은 공개 모델 시도
+    try:
+        from huggingface_hub import hf_hub_download
+        path = hf_hub_download(
+            repo_id="Kyubyong/wordvectors-ko",
+            filename="ko.bin",
+            cache_dir=str(MODELS_DIR / "word2vec"),
+        )
+        print(f"✓ 완료: {path}")
+        return path
+    except Exception as e:
+        print(f"  실패: {e}")
+        return None
+
+
+if __name__ == "__main__":
+    sroberta = download_ko_sroberta()
+    fasttext = download_korean_fasttext()
+    w2v = download_korean_word2vec()
+    print("\n=== 다운로드 결과 ===")
+    print(f"ko-sroberta: {'✓' if sroberta else '✗'}")
+    print(f"FastText: {'✓' if fasttext else '✗'}")
+    print(f"Word2Vec: {'✓' if w2v else '✗'}")