Create embeddings.py

src/embeddings.py

@@ -0,0 +1,28 @@
+import os
+import shutil
+from sentence_transformers import SentenceTransformer
+
+print("✅ embeddings.py loaded from:", __file__)
+
+# Always use a writable cache directory
+CACHE_DIR = "/tmp/huggingface"
+MODEL_NAME = "sentence-transformers/all-MiniLM-L6-v2"
+MODEL_PATH = os.path.join(CACHE_DIR, MODEL_NAME)
+
+os.environ["HF_HOME"] = CACHE_DIR
+os.environ["TRANSFORMERS_CACHE"] = CACHE_DIR
+os.environ["HF_DATASETS_CACHE"] = CACHE_DIR
+
+# If model not already cached → download once into /tmp
+if not os.path.exists(MODEL_PATH):
+    print(f"⬇️ Downloading model {MODEL_NAME} to {MODEL_PATH}")
+    _model = SentenceTransformer(MODEL_NAME, cache_folder=CACHE_DIR)
+    # Force save a copy into MODEL_PATH
+    _model.save(MODEL_PATH)
+else:
+    print(f"✅ Loading model from local path {MODEL_PATH}")
+    _model = SentenceTransformer(MODEL_PATH)
+
+def generate_embeddings(chunks: list) -> list:
+    embeddings = _model.encode(chunks, convert_to_numpy=True)
+    return embeddings.tolist()