Spaces:
Sleeping
Sleeping
Commit ·
a2f12e1
1
Parent(s): d7b3396
Disable RAG model download for instant startup (Light Mode)
Browse files- policy_engine.py +19 -32
policy_engine.py
CHANGED
|
@@ -77,41 +77,28 @@ class PolicyEngine:
|
|
| 77 |
|
| 78 |
def __init__(self):
|
| 79 |
"""Initialize FAISS index and LLM."""
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
|
|
|
| 83 |
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
snapshot_download(
|
| 87 |
-
repo_id=config.EMBEDDINGS_MODEL,
|
| 88 |
-
resume_download=True,
|
| 89 |
-
etag_timeout=60,
|
| 90 |
-
ignore_patterns=["*.onnx", "*.tflite", "*.ot", "*.h5", "*.msgpack", "*.safetensors"]
|
| 91 |
-
)
|
| 92 |
-
except Exception as e:
|
| 93 |
-
print(f"Standard download failed, retrying without hf_transfer: {e}")
|
| 94 |
-
os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "0"
|
| 95 |
-
# Fallback
|
| 96 |
-
pass
|
| 97 |
-
|
| 98 |
-
self.embeddings = HuggingFaceEmbeddings(
|
| 99 |
-
model_name=config.EMBEDDINGS_MODEL,
|
| 100 |
-
model_kwargs={'device': 'cpu'},
|
| 101 |
-
encode_kwargs={'normalize_embeddings': True}
|
| 102 |
-
)
|
| 103 |
|
| 104 |
-
|
| 105 |
-
|
| 106 |
-
str(config.FAISS_INDEX_PATH),
|
| 107 |
-
self.embeddings,
|
| 108 |
-
allow_dangerous_deserialization=True
|
| 109 |
-
)
|
| 110 |
-
except Exception as e:
|
| 111 |
-
print(f"Warning: FAISS index not found at {config.FAISS_INDEX_PATH}")
|
| 112 |
-
print(f"Error: {e}")
|
| 113 |
-
self.db = None
|
| 114 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 115 |
self.llm = ChatGroq(
|
| 116 |
model=config.LLM_MODEL,
|
| 117 |
temperature=0.5,
|
|
|
|
| 77 |
|
| 78 |
def __init__(self):
|
| 79 |
"""Initialize FAISS index and LLM."""
|
| 80 |
+
# SKIP EMBEDDINGS DOWNLOAD: "Can we not do it at all?"
|
| 81 |
+
# User requested instant startup without RAG.
|
| 82 |
+
# -----------------------------------------------------------------
|
| 83 |
+
# from huggingface_hub import snapshot_download
|
| 84 |
|
| 85 |
+
print("\n[INFO] RAG/Embeddings initialization SKIPPED by configuration.")
|
| 86 |
+
print("[INFO] Policy Engine running in 'Direct Query' mode (LLM only).")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 87 |
|
| 88 |
+
self.embeddings = None
|
| 89 |
+
self.db = None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 90 |
|
| 91 |
+
# -----------------------------------------------------------------
|
| 92 |
+
# Original logic commented out to prevent 400MB+ download on Spaces:
|
| 93 |
+
#
|
| 94 |
+
# print(f"Initializing PolicyEngine with model: {config.EMBEDDINGS_MODEL}")
|
| 95 |
+
# try:
|
| 96 |
+
# snapshot_download(...)
|
| 97 |
+
# except Exception: ...
|
| 98 |
+
# self.embeddings = HuggingFaceEmbeddings(...)
|
| 99 |
+
# self.db = FAISS.load_local(...)
|
| 100 |
+
# -----------------------------------------------------------------
|
| 101 |
+
|
| 102 |
self.llm = ChatGroq(
|
| 103 |
model=config.LLM_MODEL,
|
| 104 |
temperature=0.5,
|