Update app.py

app.py

@@ -15,7 +15,7 @@ from langchain.prompts.prompt import PromptTemplate
 from langchain.chains import ConversationalRetrievalChain
 from langchain.chains import LLMChain
 from langchain.memory import ConversationBufferMemory
-from langchain.vectorstores import Chroma
+from langchain_community.vectorstores import Chroma  # Fixed import
 from pydantic import BaseModel, Field
 from Ingestion.ingest import process_document, get_processor_for_file
 
@@ -166,40 +166,39 @@ def clean_llm_response(response):
     
     return content
 
-# Initialize LLM and Model Cache
-@st.cache_resource(experimental_allow_widgets=True)
+# Initialize LLM without widgets in the cached function
+@st.cache_resource
 def load_model():
-    with st.spinner("Loading model..."):
-        try:
-            llm = Llama.from_pretrained(
-                repo_id="stduhpf/google-gemma-3-1b-it-qat-q4_0-gguf-small",
-                filename="gemma-3-1b-it-q4_0_s.gguf",
-            )
-            return llm
-        except Exception as e:
-            st.error(f"Error loading model: {str(e)}")
-            return None
+    try:
+        llm = Llama.from_pretrained(
+            repo_id="stduhpf/google-gemma-3-1b-it-qat-q4_0-gguf-small",
+            filename="gemma-3-1b-it-q4_0_s.gguf",
+        )
+        return llm
+    except Exception as e:
+        return None
 
-# Initialize embeddings - but only when needed to avoid torch inspection issues
-@st.cache_resource(experimental_allow_widgets=True)
+# Initialize embeddings without widgets in the cached function
+@st.cache_resource
 def load_embeddings():
     from langchain_community.embeddings import HuggingFaceEmbeddings
     
-    with st.spinner("Loading embeddings..."):
-        embeddings = HuggingFaceEmbeddings(
-            model_name="sentence-transformers/all-MiniLM-L6-v2",
-            model_kwargs={'device': 'cpu'}
-        )
-        return embeddings
+    embeddings = HuggingFaceEmbeddings(
+        model_name="sentence-transformers/all-MiniLM-L6-v2",
+        model_kwargs={'device': 'cpu'}
+    )
+    return embeddings
 
 # Sidebar Configuration with improved styling
 st.sidebar.markdown("<div style='text-align: center;'><h1>🧠 DocMind AI</h1></div>", unsafe_allow_html=True)
 st.sidebar.markdown("<div style='text-align: center;'>AI-Powered Document Analysis</div>", unsafe_allow_html=True)
 st.sidebar.markdown("---")
 
-# Load LLM
+# Load LLM - Move spinner outside the cached function
 with st.sidebar:
-    llm = load_model()
+    with st.spinner("Loading model..."):
+        llm = load_model()
+    
     if llm is not None:
         st.markdown("<div class='status-success'>✅ Model loaded successfully!</div>", unsafe_allow_html=True)
     else:
@@ -539,7 +538,7 @@ def run_analysis():
             
             # Only create embeddings if we have chunks
             if all_chunks and len(all_chunks) > 0:
-                # Load embeddings
+                # Load embeddings - moving spinner outside
                 embeddings = load_embeddings()
                 
                 # Using 'None' as namespace to avoid unique ID issues with Chroma