Update app.py

app.py

@@ -61,19 +61,19 @@ A disclaimer at the bottom reminds users about potential LLM inaccuracies and th
 # Check if the LLM model is not already in the session state
 if "my_llm_model" not in st.session_state:
     # Set the default LLM model to "mistralai/Mistral-7B-Instruct-v0.3"
-    st.session_state["my_llm_model"] = "mistralai/Mistral-7B-Instruct-v0.3"
+    st.session_state['my_llm_model'] = "mistralai/Mistral-7B-Instruct-v0.3"
 # Check if the SPACE_ID environment variable is not already in the session state
 if "my_space" not in st.session_state:
-    st.session_state["my_space"] = os.environ.get("SPACE_ID") 
+    st.session_state['my_space'] = os.environ.get("SPACE_ID") 
 
 # Function to update the LLM model client
 def update_llm_model():
-    if st.session_state["my_space"]:
+    if st.session_state['my_space']:
         # Initialize the client with the model if SPACE_ID is available
-        st.session_state["client"] = InferenceClient(st.session_state["my_llm_model"])
+        st.session_state['client'] = InferenceClient(st.session_state['my_llm_model'])
     else:
         # Initialize the client with the model and token if SPACE_ID is not available
-        st.session_state["client"] = InferenceClient(st.session_state["my_llm_model"], token=os.getenv("HF_TOKEN"))
+        st.session_state['client'] = InferenceClient(st.session_state['my_llm_model'], token=os.getenv("HF_TOKEN"))
         
 # Check if the client is not already in the session state
 if "client" not in st.session_state:
@@ -82,7 +82,7 @@ if "client" not in st.session_state:
 # Check if the embeddings model is not already in the session state
 if "embeddings_model" not in st.session_state:
     # We will use the all-MiniLM-L6-v2 model for embeddings
-    st.session_state["embeddings_model"] = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
+    st.session_state['embeddings_model'] = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
 
 
 my_system_instructions = "You are a helpful assistant. Be brief and concise. Provide your answers in 100 words or less."
@@ -95,26 +95,26 @@ def delete_chat_messages():
             del st.session_state[key]
 
 def create_sentences_rag():
-    text = st.session_state["my_rag_text"]
+    text = st.session_state['my_rag_text']
 
     # Split only on .?!;: followed by space OR on \n
     pattern = r'(?<=[.?!;:])\s+|\n'
-    st.session_state["my_sentences"] = [sentence.strip() for sentence in re.split(pattern, text) if sentence.strip()]
-    sentences_ids = [i for i in range(len(st.session_state["my_sentences"]))]
+    st.session_state['my_sentences'] = [sentence.strip() for sentence in re.split(pattern, text) if sentence.strip()]
+    sentences_ids = [i for i in range(len(st.session_state['my_sentences']))]
 
     # Rolling window: include partial windows at end
-    st.session_state["my_sentences_rag_ids"] = []
-    st.session_state["my_sentences_rag"] = []
-    for rolling_window_size in range(st.session_state["min_window_size"], st.session_state["max_window_size"]+1):
-        for i in range(0, len(st.session_state["my_sentences"])-rolling_window_size+1):
-            chunk = " ".join(st.session_state["my_sentences"][i:i+rolling_window_size]).strip()
+    st.session_state['my_sentences_rag_ids'] = []
+    st.session_state['my_sentences_rag'] = []
+    for rolling_window_size in range(st.session_state['min_window_size'], st.session_state['max_window_size']+1):
+        for i in range(0, len(st.session_state['my_sentences'])-rolling_window_size+1):
+            chunk = " ".join(st.session_state['my_sentences'][i:i+rolling_window_size]).strip()
             if chunk:   
-                st.session_state["my_sentences_rag"].append(chunk)
-                st.session_state["my_sentences_rag_ids"].append(sentences_ids[i:i+rolling_window_size])
+                st.session_state['my_sentences_rag'].append(chunk)
+                st.session_state['my_sentences_rag_ids'].append(sentences_ids[i:i+rolling_window_size])
                 # print(f"*****{chunk}*****\n")
 
-    print(len(st.session_state["my_sentences_rag"]))
-    st.session_state["my_embeddings"] = st.session_state["embeddings_model"].encode(st.session_state["my_sentences_rag"])
+    print(len(st.session_state['my_sentences_rag']))
+    st.session_state['my_embeddings'] = st.session_state['embeddings_model'].encode(st.session_state['my_sentences_rag'])
 
 
 augmented_prompt = ""
@@ -137,9 +137,9 @@ By using the Software, you agree to the terms and conditions of the disclaimer."
 
     # Add a selectbox for model selection
     st.selectbox("Select the model to use:",
-                ["mistralai/Mistral-7B-Instruct-v0.3",
-                 "Qwen/Qwen2.5-72B-Instruct", 
-                "HuggingFaceH4/zephyr-7b-beta"], 
+                ['mistralai/Mistral-7B-Instruct-v0.3',
+                 'Qwen/Qwen2.5-72B-Instruct', 
+                'HuggingFaceH4/zephyr-7b-beta'], 
                 key="my_llm_model", on_change=update_llm_model)
     
     # Add a text are for the system instructions
@@ -171,9 +171,9 @@ By using the Software, you agree to the terms and conditions of the disclaimer."
 # Check if the chat messages are not already in the session state
 if "my_chat_messages" not in st.session_state:
     # Initialize the chat messages list in the session state
-    st.session_state["my_chat_messages"] = []
+    st.session_state['my_chat_messages'] = []
     # Add the system instructions to the chat messages
-    st.session_state["my_chat_messages"].append({"role": "system", "content": st.session_state["my_system_instructions"]})
+    st.session_state['my_chat_messages'].append({"role": "system", "content": st.session_state['my_system_instructions']})
 
 
 # print(100*"-")
@@ -190,32 +190,32 @@ with column_2:
     messages_container.chat_message("ai", avatar=":material/robot_2:").markdown(first_message)
     
     # Iterate through the chat messages stored in the session state
-    for message in st.session_state["my_chat_messages"]:
-        if message["role"] == "user":
+    for message in st.session_state['my_chat_messages']:
+        if message['role'] == "user":
             # Display user messages with a specific avatar - https://fonts.google.com/icons
-            messages_container.chat_message(message["role"], avatar=":material/psychology_alt:").markdown(message["content"])
-        elif message["role"] == "assistant":
+            messages_container.chat_message(message['role'], avatar=":material/psychology_alt:").markdown(message['content'])
+        elif message['role'] == "assistant":
             # Display assistant messages with a specific avatar
-            messages_container.chat_message(message["role"], avatar=":material/robot_2:").markdown(message["content"])
+            messages_container.chat_message(message['role'], avatar=":material/robot_2:").markdown(message['content'])
 
     # Check if there is a new prompt from the user
     if prompt := st.chat_input("you may ask here your questions"):
         
         split_prompt = prompt.split(" ")
         all_sub_prompts = []
-        for jj in range(st.session_state["nof_min_sub_prompts"], st.session_state["nof_max_sub_prompts"]+1):
+        for jj in range(st.session_state['nof_min_sub_prompts'], st.session_state['nof_max_sub_prompts']+1):
             for ii in range(len(split_prompt)):
                 i_split = " ".join(split_prompt[ii:ii+jj]).strip()
                 if i_split:
                     all_sub_prompts.append(i_split)
         
-        similarities_to_question = np.zeros(len(st.session_state["my_embeddings"]))
+        similarities_to_question = np.zeros(len(st.session_state['my_embeddings']))
         for sub_prompt in all_sub_prompts:
             # Encode the user's prompt to get its embedding
             my_question_embedding = st.session_state.embeddings_model.encode([sub_prompt])
             
             # Calculate the cosine similarity between the prompt embedding and stored embeddings
-            similarities_to_question += cosine_similarity(my_question_embedding, st.session_state["my_embeddings"]).flatten()
+            similarities_to_question += cosine_similarity(my_question_embedding, st.session_state['my_embeddings']).flatten()
         similarities_to_question /= len(all_sub_prompts)
         
         # Get the indices of the top similar sentences
@@ -223,24 +223,24 @@ with column_2:
         sorted_indices_rag = similarities_to_question.argsort()[::-1]
         sorted_indices_sentences = []
         max_similarity = 0
-        # for irag in range(st.session_state["nof_keep_sentences"]):
+        # for irag in range(st.session_state['nof_keep_sentences']):
         irag = 0
-        while len(set(sorted_indices_sentences))<st.session_state["nof_keep_sentences"]:
-            sorted_indices_sentences.extend(st.session_state["my_sentences_rag_ids"][sorted_indices_rag[irag]])
+        while len(set(sorted_indices_sentences))<st.session_state['nof_keep_sentences']:
+            sorted_indices_sentences.extend(st.session_state['my_sentences_rag_ids'][sorted_indices_rag[irag]])
             max_similarity = max(max_similarity, similarities_to_question[sorted_indices_rag[irag]])
             with bottom_col1:
-                str_conf = f"Confidence: {similarities_to_question[sorted_indices_rag[irag]]:.5f}, Sentences IDs: {st.session_state["my_sentences_rag_ids"][sorted_indices_rag[irag]]}"
+                str_conf = f"Confidence: {similarities_to_question[sorted_indices_rag[irag]]:.5f}, Sentences IDs: {st.session_state['my_sentences_rag_ids'][sorted_indices_rag[irag]]}"
                 with st.expander(f"Chunk: {str(irag+1)} {str_conf}"):
-                    for idx in st.session_state["my_sentences_rag_ids"][sorted_indices_rag[irag]]:
-                        st.write(f"{st.session_state["my_sentences"][idx]}")
+                    for idx in st.session_state['my_sentences_rag_ids'][sorted_indices_rag[irag]]:
+                        st.write(f"{st.session_state['my_sentences'][idx]}")
             irag += 1
 
         sorted_indices_sentences = sorted(list(set(sorted_indices_sentences)))
 
         # Construct the augmented prompt with the similar sentences
-        if max_similarity > st.session_state["my_similarity_threshold"]:
+        if max_similarity > st.session_state['my_similarity_threshold']:
             augmented_prompt = "This is my context:" + "\n\n" + 20*"-" + "\n\n" 
-            augmented_prompt += "\n".join([st.session_state["my_sentences"][idx] for idx in sorted_indices_sentences])
+            augmented_prompt += "\n".join([st.session_state['my_sentences'][idx] for idx in sorted_indices_sentences])
             augmented_prompt += "\n\n" + 20*"-" + "\n\n" + "If the above context is not relevant to the prompt, ignore the context and reply based only on the prompt."
             augmented_prompt += "\n\n" + 20*"-" + "\n\n" + "If the above context is relevant to the prompt, reply based on the context and the prompt."
             augmented_prompt += "\n\n" + 20*"-" + "\n\n" + "The prompt is:"
@@ -255,28 +255,26 @@ with column_2:
         # Display the user's prompt in the chat container with a specific avatar
         messages_container.chat_message("user", avatar=":material/psychology_alt:").markdown(prompt) 
         # Append the augmented prompt to the chat messages in the session state
-        st.session_state["my_chat_messages"].append({"role": "user", "content": augmented_prompt})
+        st.session_state['my_chat_messages'].append({"role": "user", "content": augmented_prompt})
         # Create an empty container for the streaming response from the assistant
         with messages_container.chat_message("ai", avatar=":material/robot_2:"):
             response_placeholder = st.empty()
             response = ""
             # Stream the response from the assistant and update the placeholder
-            for chunk in st.session_state["client"].chat.completions.create(messages=st.session_state["my_chat_messages"], stream=True, max_tokens=512):
+            for chunk in st.session_state['client'].chat.completions.create(messages=st.session_state['my_chat_messages'], stream=True, max_tokens=512):
                 if chunk.choices[0].delta.content:
                     response += chunk.choices[0].delta.content
                     # Use markdown to update the response placeholder with the streamed content
                     response_placeholder.markdown(response)
 
         # Remove the last message from the chat messages in the session state
-        st.session_state["my_chat_messages"].pop()
+        st.session_state['my_chat_messages'].pop()
         # Append the user's original prompt to the chat messages in the session state
-        st.session_state["my_chat_messages"].append({"role": "user", "content": prompt})
+        st.session_state['my_chat_messages'].append({"role": "user", "content": prompt})
         # Append the assistant's response to the chat messages in the session state
-        st.session_state["my_chat_messages"].append({"role": "assistant", "content": response})
+        st.session_state['my_chat_messages'].append({"role": "assistant", "content": response})
 
         with bottom_col2:
             # Display the chat messages history
             st.write("Messages History All:")
-            st.json(st.session_state["my_chat_messages"], expanded=False)
-
-
+            st.json(st.session_state['my_chat_messages'], expanded=False)