be a bit more conservative when summarizing in-line

matrix_pygmalion_bot/bot/ai/langchain.py

@@ -243,8 +243,8 @@ class AI(object):
             tmp_prompt_text = prompt.format(chat_history=conversation_memory.buffer, input=message.content)
             prompt_len = self.llm_chat.get_num_tokens(tmp_prompt_text)
 
-            if prompt_len+256 > 2000:
-                logger.warning(f"Prompt too large. Estimated {prompt_len} tokens")
+            if prompt_len+200 > 2048:
+                logger.warning(f"Prompt too large. Estimated {prompt_len} tokens. Summarizing...")
                 await reply_fn(f"<WARNING> Prompt too large. Estimated {prompt_len} tokens")
                 if i == 0:
                     await conversation_memory.prune_memory(conversation_memory.min_len)