Merge branch 'main' into feat-add-guildid

src/discord_bot/bot.py

@@ -21,7 +21,7 @@ async def on_ready() -> None:
 
 @interactions.listen()
 async def on_message_create(event: interactions.api.events.MessageCreate) -> None:
-  print(f"message received: {event.message.content}")
+  print(f"Message received: {event.message.content}")
 
 
 @interactions.slash_command(name="ask", description="Ask an LLM", scopes=[DISCORD_GUILD_ID])
@@ -43,7 +43,8 @@ async def ask_model(ctx: interactions.SlashContext, model: str = "", prompt: str
   await ctx.defer()
 
   response = await answer_question(model, prompt, AI_SERVER_URL)
-  await ctx.send(response)
+  for r in response:
+    await ctx.send(r)
 
 
 @ask_model.autocomplete("model")

src/discord_bot/llm.py

@@ -1,21 +1,72 @@
 """
 Although this module uses `openai` package but we are routing it
-through our LiteLLM proxy to interact with Ollama and OpenAI models 
+through our LiteLLM proxy to interact with Ollama and OpenAI models
 by modifying the `base_url`.
 """
 
 import openai
 
+MAX_CHARACTERS = 2000
+QUESTION_CUT_OFF_LENGTH = 150
+RESERVED_SPACE = 50  # for other additional strings. E.g. number `(1/4)`, `Q: `, `A: `, etc.
 
-async def answer_question(model: str, question: str, server_url: str) -> str:
+
+async def answer_question(model: str, question: str, server_url: str) -> list[str]:
   try:
     client = openai.AsyncOpenAI(base_url=server_url, api_key="FAKE")
     response = await client.chat.completions.create(
       model=model,
       messages=[{"role": "user", "content": question}],
     )
-    out = response.choices[0].message.content or "No response from the model. Please try again"
-    return out
+    content = response.choices[0].message.content or "No response from the model. Please try again"
+    messages = split(content)
+    messages = add_number(messages)
+    messages = add_question(messages, question)
+    return messages
 
   except Exception as e:
-    return f"Error: {e}"
+    return split(f"Error: {e}")
+
+
+def split(answer: str) -> list[str]:
+  """
+  Split the answer into a list of smaller strings so that
+  each element is less than MAX_CHARACTERS characters.
+  Full sentences are preserved.
+  """
+  limit = MAX_CHARACTERS - RESERVED_SPACE - QUESTION_CUT_OFF_LENGTH
+  messages = []
+  answer = answer.strip()
+
+  while len(answer) > limit:
+    last_period = answer[:limit].rfind(".")
+    if last_period == -1:
+      last_period = answer[:limit].rfind(" ")
+    messages.append(answer[: last_period + 1])
+    answer = answer[last_period + 1 :]
+
+  messages.append(answer)
+
+  return messages
+
+
+def add_question(messages: list[str], questions: str) -> list[str]:
+  """
+  Add the asked question to the beginning of each message.
+  """
+  return [(f"Q: {questions[:QUESTION_CUT_OFF_LENGTH]}\n" + f"A: {message}") for message in messages]
+
+
+def add_number(messages: list[str]) -> list[str]:
+  """
+  Add the number to the beginning of each message. E.g. `(1/4)`
+  Do nothing if the length of `messages` is 1.
+  """
+  if len(messages) == 1:
+    return messages
+
+  for i, message in enumerate(messages):
+    message = message.strip()
+    messages[i] = f"({i+1}/{len(messages)}) {message}"
+
+  return messages