Review comments

alphastats/gui/pages/06_LLM.py

@@ -235,7 +235,7 @@ def llm_config():
         )
 
         with st.spinner("Processing initial prompt..."):
-            llm_integration.chat_completion(initial_prompt, pinned=True)
+            llm_integration.chat_completion(initial_prompt, pin_message=True)
 
         st.rerun(scope="app")
     except AuthenticationError:
@@ -249,7 +249,7 @@ def llm_config():
 def llm_chat(
     llm_integration: LLMIntegration,
     show_all: bool = False,
-    show_inidvidual_tokens: bool = False,
+    show_individual_tokens: bool = False,
 ):
     """The chat interface for the LLM analysis."""
 
@@ -269,11 +269,11 @@ def llm_chat(
                 total_tokens += tokens
             if message[MessageKeys.PINNED]:
                 pinned_tokens += tokens
-            if message[MessageKeys.PINNED] or show_inidvidual_tokens:
+            if message[MessageKeys.PINNED] or show_individual_tokens:
                 token_message = ""
                 if message[MessageKeys.PINNED]:
                     token_message += ":pushpin: "
-                if show_inidvidual_tokens:
+                if show_individual_tokens:
                     token_message += f"*estimated tokens: {str(tokens)}*"
                 st.markdown(token_message)
             if not message[MessageKeys.IN_CONTEXT]:

alphastats/llm/llm_integration.py

@@ -199,6 +199,7 @@ def estimate_tokens(
                 ]
             )
         except KeyError:
+            # if the model is not in the tiktoken library a key error is raised by encoding_for_model, we use a rough estimate instead
             total_tokens = sum(
                 [
                     len(message[MessageKeys.CONTENT]) / average_chars_per_token
@@ -224,32 +225,36 @@ def _truncate_conversation_history(
         # TODO: avoid important messages being removed (e.g. facts about genes)
         # TODO: find out how messages can be None type and handle them earlier
         total_tokens = self.estimate_tokens(self._messages, average_chars_per_token)
-        while total_tokens > self._max_tokens and len(self._messages) > 1:
-            oldest_unpinned = -1
+        while total_tokens > self._max_tokens:
+            if len(self._messages) == 1:
+                raise ValueError(
+                    "Truncating conversation history failed, as the only remaining message exceeds the token limit. Please increase the token limit and reset the LLM analysis."
+                )
+            oldest_not_pinned = -1
             for message_idx, message in enumerate(self._messages):
                 if not message[MessageKeys.PINNED]:
-                    oldest_unpinned = message_idx
+                    oldest_not_pinned = message_idx
                     break
-            if oldest_unpinned == -1:
+            if oldest_not_pinned == -1:
                 raise ValueError(
                     "Truncating conversation history failed, as all remaining messages are pinned. Please increase the token limit and reset the LLM analysis, or unpin messages."
                 )
-            removed_message = self._messages.pop(oldest_unpinned)
+            removed_message = self._messages.pop(oldest_not_pinned)
             warnings.warn(
-                f"Truncating conversation history to stay within token limits.\nRemoved message:{removed_message[MessageKeys.ROLE]}: {removed_message[MessageKeys.CONTENT][0:min(30, len(removed_message[MessageKeys.CONTENT]))]}..."
+                f"Truncating conversation history to stay within token limits.\nRemoved message:{removed_message[MessageKeys.ROLE]}: {removed_message[MessageKeys.CONTENT][0:30]}..."
             )
             while (
                 removed_message[MessageKeys.ROLE] == Roles.ASSISTANT
-                and self._messages[oldest_unpinned][MessageKeys.ROLE] == Roles.TOOL
+                and self._messages[oldest_not_pinned][MessageKeys.ROLE] == Roles.TOOL
             ):
                 # This is required as the chat completion fails if there are tool outputs without corresponding tool calls in the message history.
-                removed_toolmessage = self._messages.pop(oldest_unpinned)
+                removed_toolmessage = self._messages.pop(oldest_not_pinned)
                 warnings.warn(
-                    f"Removing corresponsing tool output as well.\nRemoved message:{removed_toolmessage[MessageKeys.ROLE]}: {removed_toolmessage[MessageKeys.CONTENT][0:min(30, len(removed_toolmessage[MessageKeys.CONTENT]))]}..."
+                    f"Removing corresponsing tool output as well.\nRemoved message:{removed_toolmessage[MessageKeys.ROLE]}: {removed_toolmessage[MessageKeys.CONTENT][0:30]}..."
                 )
                 if len(self._messages) == 0:
                     raise ValueError(
-                        "Truncating conversation history failed, as the most recent artifacts exceeded the token limit. Please increase the token limit and reset the LLM analysis."
+                        "Truncating conversation history failed, as the artifact from the last call exceeds the token limit. Please increase the token limit and reset the LLM analysis."
                     )
             total_tokens = self.estimate_tokens(self._messages, average_chars_per_token)
 
@@ -407,7 +412,7 @@ def get_chat_log_txt(self) -> str:
         return chatlog
 
     def chat_completion(
-        self, prompt: str, role: str = Roles.USER, *, pinned=False
+        self, prompt: str, role: str = Roles.USER, *, pin_message=False
     ) -> None:
         """
         Generate a chat completion based on the given prompt and manage any resulting artifacts.
@@ -418,15 +423,15 @@ def chat_completion(
             The user's input prompt
         role : str, optional
             The role of the message sender, by default "user"
-        pinned : bool, optional
+        pin_message : bool, optional
             Whether the prompt and assistant reply should be pinned, by default False
 
         Returns
         -------
         Tuple[str, Dict[str, Any]]
             A tuple containing the generated response and a dictionary of new artifacts
         """
-        self._append_message(role, prompt, pinned=pinned)
+        self._append_message(role, prompt, pinned=pin_message)
 
         try:
             response = self._chat_completion_create()
@@ -441,7 +446,7 @@ def chat_completion(
 
                 content, _ = self._handle_function_calls(tool_calls)
 
-            self._append_message(Roles.ASSISTANT, content, pinned=pinned)
+            self._append_message(Roles.ASSISTANT, content, pinned=pin_message)
 
         except ArithmeticError as e:
             error_message = f"Error in chat completion: {str(e)}"