Update app.py

app.py

@@ -1,6 +1,7 @@
 # app.py - Hybrid Memory-Augmented Graph RAG with Groq
 import hashlib
 import os
+import pickle
 from collections import defaultdict
 from datetime import datetime
 
@@ -37,7 +38,11 @@ def rrf_score(rank: int, k: int = RRF_K) -> float:
 # ========================= MEMORY LAYER =========================
 class HybridMemory:
     def __init__(self):
-        self.embeddings = HuggingFaceEmbeddings(model_name=EMBED_MODEL_NAME)
+        self.embeddings = HuggingFaceEmbeddings(
+            model_name=EMBED_MODEL_NAME,
+            cache_folder="/tmp/model_cache",
+            model_kwargs={"device": "cpu"},
+        )
         self.vector_memory = None
         self.graph_memory = nx.Graph()
         self.memory_dir = "user_memory"
@@ -46,17 +51,19 @@ class HybridMemory:
 
     def load_memory(self):
         vector_path = os.path.join(self.memory_dir, "vector_memory")
-        graph_path = os.path.join(self.memory_dir, "memory_graph.gpickle")
-        
+        graph_path = os.path.join(self.memory_dir, "memory_graph.pkl")
+
         if os.path.exists(vector_path):
             self.vector_memory = FAISS.load_local(vector_path, self.embeddings, allow_dangerous_deserialization=True)
         if os.path.exists(graph_path):
-            self.graph_memory = nx.read_gpickle(graph_path)
+            with open(graph_path, "rb") as f:
+                self.graph_memory = pickle.load(f)
 
     def save_memory(self):
         if self.vector_memory:
             self.vector_memory.save_local(os.path.join(self.memory_dir, "vector_memory"))
-        nx.write_gpickle(self.graph_memory, os.path.join(self.memory_dir, "memory_graph.gpickle"))
+        with open(os.path.join(self.memory_dir, "memory_graph.pkl"), "wb") as f:
+            pickle.dump(self.graph_memory, f)
 
     def add_memory(self, text: str, memory_type: str = "fact", metadata: dict = None):
         if metadata is None:
@@ -110,11 +117,21 @@ class HybridMemory:
 
 # ========================= ORIGINAL PIPELINE FUNCTIONS =========================
 @st.cache_resource
-def load_embed_model(): return SentenceTransformer(EMBED_MODEL_NAME)
+def load_embed_model():
+    return SentenceTransformer(EMBED_MODEL_NAME, cache_folder="/tmp/model_cache")
+
 @st.cache_resource
-def load_reranker(): return CrossEncoder(RERANKER_MODEL)
+def load_reranker():
+    return CrossEncoder(RERANKER_MODEL)
+
 @st.cache_resource
-def load_lc_embeddings(): return HuggingFaceEmbeddings(model_name=EMBED_MODEL_NAME)
+def load_lc_embeddings():
+    return HuggingFaceEmbeddings(
+        model_name=EMBED_MODEL_NAME,
+        cache_folder="/tmp/model_cache",
+        model_kwargs={"device": "cpu"},
+    )
+
 @st.cache_resource
 def load_spacy():
     try: