Commit a268126
Changed files (5)
src
src/llm/gemini.py
@@ -55,6 +55,7 @@ async def gemini_response(
disable_thinking: bool = False,
include_thoughts: bool = True,
system_prompt: str | None = None,
+ silent: bool = False,
**kwargs,
) -> dict:
r"""Get Gemini response.
@@ -64,9 +65,12 @@ async def gemini_response(
message (Message): The trigger message object.
conversations (list[Message]): list of chat conversations.
modality (str): response modality
+ enable_tools (bool, optional): Whether to enable tools. Defaults to True.
append_grounding (bool, optional): Whether to append grounding to the response. Defaults to True.
disable_thinking (bool, optional): Whether to disable thinking. Defaults to False.
include_thoughts (bool, optional): Whether to include thoughts. Defaults to True.
+ system_prompt (str | None, optional): System prompt. Defaults to None.
+ silent (bool, optional): Whether to disable progressing. Defaults to False.
"""
info = parse_msg(message)
model = GEMINI.TEXT_MODEL if modality == "text" else GEMINI.IMG_MODEL
@@ -84,8 +88,8 @@ async def gemini_response(
try:
real_prompt = clean_cmd_prefix(info["text"]) or clean_cmd_prefix(info["reply_text"])
msg = f"🤖**{model_name}**: 思考中...\n👤**[{info['full_name'] or info['ctitle']}](tg://user?id={info['uid']})**: “{real_prompt}”"[:TEXT_LENGTH]
- status_msg = (await send2tg(client, message, texts=msg, **kwargs))[0]
- kwargs["progress"] = status_msg
+ if not silent and kwargs.get("show_progress"):
+ kwargs["progress"] = (await send2tg(client, message, texts=msg, **kwargs))[0]
genconfig |= {"response_modalities": response_modalities}
if enable_tools and modality == "text":
genconfig |= {"tools": tools}
@@ -101,7 +105,7 @@ async def gemini_response(
logger.trace(params)
if modality == "image":
return await gemini_nonstream(client, message, model_name, params, clean_marks=True, append_grounding=append_grounding, **kwargs)
- return await gemini_stream(client, message, model_name, params, append_grounding=append_grounding, **kwargs)
+ return await gemini_stream(client, message, model_name, params, append_grounding=append_grounding, silent=silent, **kwargs)
except Exception as e:
logger.error(e)
return {}
src/llm/summary.py
@@ -7,20 +7,19 @@ from datetime import datetime, timedelta
from zoneinfo import ZoneInfo
from loguru import logger
-from openai import DefaultAsyncHttpxClient
from pyrogram.client import Client
from pyrogram.types import Chat, Message
+from pyrogram.types.messages_and_media.message import Str
-from config import GPT, MAX_MESSAGE_SUMMARY, PREFIX, PROXY, TID, TZ
-from llm.prompts import refine_prompts
-from llm.response import send_to_gpt
-from llm.utils import count_tokens, sample_key
-from messages.chat_history import get_parsed_chat_history
+from config import GPT, MAX_MESSAGE_SUMMARY, PREFIX, TID, TZ
+from llm.gpt import gpt_response
+from llm.utils import BOT_TIPS, count_tokens
+from messages.chat_history import get_history_info_list
from messages.parser import parse_msg
from messages.progress import modify_progress
from messages.sender import send2tg
from messages.utils import equal_prefix, startswith_prefix, to_int
-from utils import nowdt
+from utils import nowdt, rand_number
HELP = f"""🤖**AI总结历史消息** (最多{MAX_MESSAGE_SUMMARY}条)
⚠️使用`{PREFIX.COMBINATION}`命令只生成聊天记录文件, 不进行AI总结
@@ -59,6 +58,36 @@ HELP = f"""🤖**AI总结历史消息** (最多{MAX_MESSAGE_SUMMARY}条)
- 用上述各种`{PREFIX.AI_SUMMARY}`命令回复消息M, 视为将截止时间设为消息M的发送时间
- 如果用户名中有空格, 请去除空格。例如: 想指定用户为John Doe请使用 `@JohnDoe`
"""
+
+SYSTEM_PROMPT = """总结以下网络聊天记录, 识别关键主题、争议话题以及重要观点。提供一个简明的总结, 保留原始意图和上下文。如有必要, 引用原始用户名和时间戳。
+每一条消息的格式如下:
+{
+ "username": "消息发送者",
+ "time": "消息发送时间",
+ "url": "消息链接",
+ "message": "本条消息内容",
+ "reply_to_message": "被此条消息回复的消息"
+}
+
+# 步骤
+1. 阅读聊天记录: 仔细查看对话内容, 了解讨论的流程和上下文。
+2. 识别关键主题: 提取整个聊天中讨论的主要话题。
+3. 忽略废话及无关内容, 专注于关键信息。
+4. 突出争议话题: 记录任何分歧或意见不同的地方。
+5. 识别重要观点: 捕捉参与者提出的重要观点或论点。
+6. 保留意图和上下文: 确保总结反映对话的原始意义和上下文。
+7. 引用用户名和时间戳: 在适当情况下, 引用用户名和时间戳以为某些陈述提供上下文。
+8. 撰写总结: 以简洁的语言编写总结, 同时包含必要的引用。
+
+# 输出格式
+- 使用中文撰写总结。
+- 简明扼要地总结聊天记录的内容。
+- 在必要时引用用户名和时间。
+- 保持清晰和简洁的表达。
+- 引用用户名时, 请使用 **username** 格式。如: **username**
+- 引用时间时, 请使用 [HH:MM:SS](url) 格式。如: [12:30:00](https://t.me/c/1234/56789)
+"""
+
DAILY_SUMMARY_PREFIX = "🏪**#爬楼助手**\n"
CONTEXT_FILENAME = "聊天记录.txt"
@@ -122,126 +151,96 @@ async def ai_summary(client: Client, message: Message, summary_prefix: str | Non
if kwargs.get("show_progress") and "progress" not in kwargs:
res = await send2tg(client, message, texts=f"📝正在获取历史消息...\n⏩开始时间: {begin_time:%m-%d %H:%M:%S}\n⏯️结束时间: {end_time:%m-%d %H:%M:%S}", **kwargs)
kwargs["progress"] = res[0]
- history = await get_parsed_chat_history(client, info["cid"], offset_id, num_history, begin_time, end_time, filter_users)
+ history_list = await get_history_info_list(client, info["cid"], offset_id, num_history, begin_time, end_time, filter_users)
# parse the history contexts
- parsed = await get_contexts(history)
- if not parsed["txt_format"]:
+ parsed = await parse_history_list(history_list)
+ if parsed["num_message"] == 0:
await send2tg(client, message, texts=f"{num_history}条历史消息中未找到符合条件的消息", **kwargs)
await modify_progress(del_status=True, **kwargs)
return
- contexts = refine_prompts(parsed["system_context"] + [{"role": "user", "content": parsed["user_context"]}])
- sysmtem_tokens = count_tokens(contexts[0]["content"])
- user_tokens = count_tokens(contexts[-1]["content"])
- total_tokens = sysmtem_tokens + user_tokens
- summary_model = GPT.SUMMARY_MODEL
- summary_model_name = GPT.SUMMARY_MODEL_NAME
- max_tokens = int(GPT.SUMMARY_MODEL_MAX_OUTPUT_LENGTH)
+ num_tokens = count_tokens(SYSTEM_PROMPT + parsed["history"])
msg = f"⏩开始时间: {parsed['begin_time']:%m-%d %H:%M:%S}\n"
msg += f"⏯️结束时间: {parsed['end_time']:%m-%d %H:%M:%S}\n"
- msg += f"🔢消息条数: {len(parsed['user_context'])}\n"
- msg += f"🔠Token数: {total_tokens}"
+ msg += f"🔢消息条数: {parsed['num_message']}\n"
+ msg += f"🔠Token数: {num_tokens}"
# send contexts as txt file
with io.BytesIO(parsed["txt_format"].encode("utf-8")) as f:
await client.send_document(to_int(message.chat.id), f, file_name=CONTEXT_FILENAME, caption=msg)
if not need_summay:
await modify_progress(del_status=True, **kwargs)
return
- await modify_progress(text=f"🤖**{summary_model_name}**总结中...\n{msg}", force_update=True, **kwargs)
- config = {
- "friendly_name": summary_model_name,
- "client": {"api_key": sample_key(GPT.SUMMARY_API_KEY), "base_url": GPT.SUMMARY_BASE_URL, "http_client": DefaultAsyncHttpxClient(proxy=PROXY.GPT)},
- "completions": {"model": summary_model},
- }
- config["completions"]["messages"] = contexts
- # set max_tokens for the model
- if "o1" in summary_model or "o3" in summary_model: # o1 or newer models use `max_completion_tokens`
- config["completions"]["max_completion_tokens"] = max_tokens
- else:
- config["completions"]["max_tokens"] = max_tokens
- config["client"]["timeout"] = int(GPT.SUMMARY_TIMEOUT)
- response = await send_to_gpt(config, **kwargs)
- if texts := response.get("content"):
- texts = texts.strip("`")
+ await modify_progress(text=f"🤖AI总结中...\n{msg}", force_update=True, **kwargs)
+ # Construct a message to call GPT
+ ai_msg = Message(
+ id=rand_number(),
+ chat=message.chat,
+ text=Str(GPT.SUMMARY_CMD),
+ reply_to_message=Message(id=rand_number(), chat=message.chat, text=Str(parsed["history"])),
+ )
+ response = await gpt_response(
+ client,
+ ai_msg,
+ system_prompt=SYSTEM_PROMPT,
+ enable_gpt_tools=False,
+ include_thoughts=False,
+ append_grounding=False,
+ silent=True,
+ )
+ if texts := response.get("texts"):
if summary_prefix is None:
- summary_prefix = f"🤖**{summary_model_name}**:\n"
+ model_name = response.get("model_name", "AI总结")
+ summary_prefix = f"🤖**{model_name}**:\n"
kwargs["reply_msg_id"] = -1 # DO NOT send as a reply message
await send2tg(client, message, texts=f"{summary_prefix}⏩开始时间: {begin_time:%m-%d %H:%M:%S}\n⏯️结束时间: {end_time:%m-%d %H:%M:%S}\n{texts}", **kwargs)
await modify_progress(del_status=True, **kwargs)
-async def get_contexts(history: list[dict]) -> dict:
- """Get GPT contexts based on parsed chat history.
+async def parse_history_list(info_list: list[dict]) -> dict:
+ """Parse chat history info list.
Currently, we only summarize text contents.
"""
- system_context = [
- {
- "role": "system", # system prompt
- "content": [
- {
- "type": "text",
- "text": """总结以下网络聊天记录, 识别关键主题、争议话题以及重要观点。提供一个简明的总结, 保留原始意图和上下文。如有必要, 引用原始用户名和时间戳, 并使用清晰的语言。
-每一条消息的格式如下:
-{
- "time": 消息发送时间,
- "url": 消息链接,
- "username": 消息发送者,
- "message": 本条消息内容,
- "reply_to_message": 回复消息的原始内容, 如果本消息并不回复其他消息, 则不存在该字段
-}
-
-# 步骤
-1. 阅读聊天记录: 仔细查看对话内容, 了解讨论的流程和上下文。
-2. 识别关键主题: 提取整个聊天中讨论的主要话题。
-3. 忽略废话及无关内容, 专注于关键信息。
-4. 突出争议话题: 记录任何分歧或意见不同的地方。
-5. 识别重要观点: 捕捉参与者提出的重要观点或论点。
-6. 保留意图和上下文: 确保总结反映对话的原始意义和上下文。
-7. 引用用户名和时间戳: 在适当情况下, 引用用户名和时间戳以为某些陈述提供上下文。
-8. 撰写总结: 以简洁的语言编写总结, 同时包含必要的引用。
-
-# 输出格式
-- 使用中文撰写总结。
-- 简明扼要地总结聊天记录的内容。
-- 在必要时引用用户名和时间。
-- 保持清晰和简洁的表达。
-- 引用用户名时, 请使用 **username** 格式。如: **username**
-- 引用时间时, 请使用 [HH:MM:SS](url) 格式。如: [12:30:00](https://t.me/c/1234/56789)
-""",
- }
- ],
- }
- ]
- user_context = []
- for info in history:
+ messages: list[dict] = [] # hold user messages
+ for info in info_list:
if info["file_name"] == CONTEXT_FILENAME:
continue
if info["is_bot"]: # bots
continue
if info["text"]: # currently, we only include texts
- if len(user_context) == 0:
+ if len(messages) == 0:
begin_time = info["datetime"]
end_time = info["datetime"]
media_type = f"[{info['mtype']}] " if info["mtype"] != "text" else ""
content = {
+ "username": info["full_name"],
"time": f"{info['datetime']:%H:%M:%S}",
"url": info["message_url"],
- "username": info["full_name"],
"message": media_type + info["text"],
}
- if reply_msg_content := get_message_by_id(history, info.get("reply_to_message_id")):
+ if reply_msg_content := get_message_by_id(info_list, info.get("reply_to_message_id")):
content["reply_to_message"] = reply_msg_content
- user_context.append({"type": "text", "text": str(content)})
- if not user_context:
+ messages.append(content)
+ if not messages:
return {}
- return {"system_context": system_context, "user_context": user_context, "txt_format": get_txt_format(history), "begin_time": begin_time, "end_time": end_time}
+ history = json.dumps(messages, ensure_ascii=False)
+ """IMPORTANT: We need to remove `BOT_TIPS` in the history!
-def get_txt_format(history: list[dict]) -> str:
+ Because we need to call `gpt_response` function,
+ it uses `BOT_TIPS` to check if the message is from GPT model.
+
+ If the history contains `BOT_TIPS`, the context of this message will be `model` (not `user`)
+ But `model` only message is not allowed, so we need to remove `BOT_TIPS`
+ """
+ history = history.replace(BOT_TIPS, "")
+ return {"history": history, "num_message": len(messages), "txt_format": get_txt_format(info_list), "begin_time": begin_time, "end_time": end_time}
+
+
+def get_txt_format(info_list: list[dict]) -> str:
"""Format the history as plaintext."""
txt_format = ""
txt_mediagroup_ids = set() # record processed mediagroup messages
- for info in history:
+ for info in info_list:
if info["file_name"] == CONTEXT_FILENAME:
continue
if info["media_group_id"] in txt_mediagroup_ids:
@@ -251,40 +250,39 @@ def get_txt_format(history: list[dict]) -> str:
if info["mtype"] != "text": # not plaintext message
# media group
if info["media_group_id"] > 0:
- media_types = [f"[{x['mtype']}]" for x in history if x["media_group_id"] == info["media_group_id"]]
+ media_types = [f"[{x['mtype']}]" for x in info_list if x["media_group_id"] == info["media_group_id"]]
txt_format += " ".join(media_types)
txt_mediagroup_ids.add(info["media_group_id"])
else:
txt_format += f"[{info['mtype']}]"
txt_format += info["text"]
# append quote msg
- reply_msg_content = get_message_by_id(history, info.get("reply_to_message_id"))
- if reply_msg_content:
+ if reply_msg_content := get_message_by_id(info_list, info.get("reply_to_message_id")):
txt_format += f"\n<quote>{reply_msg_content['username']}: {reply_msg_content['message']}</quote>"
txt_format += "\n\n"
return txt_format
-def get_message_by_id(history: list[dict], message_id: int | None = None) -> dict:
+def get_message_by_id(info_list: list[dict], message_id: int | None = None) -> dict:
"""Get message by id."""
if not message_id:
return {}
- info = next((info for info in history if info["mid"] == message_id), {})
+ info = next((info for info in info_list if info["mid"] == message_id), {})
if not info:
return {}
media_type = f"[{info['mtype']}] " if info["mtype"] != "text" else ""
return {
+ "username": info["full_name"],
"time": f"{info['datetime']:%H:%M:%S}",
"url": info["message_url"],
- "username": info["full_name"],
"message": media_type + info["text"],
}
-def get_media_group_by_id(history: list[dict], media_group_id: int | None = None) -> list[dict]:
+def get_media_group_by_id(info_list: list[dict], media_group_id: int | None = None) -> list[dict]:
if not media_group_id:
return []
- return [x for x in history if x["media_group_id"] == media_group_id]
+ return [x for x in info_list if x["media_group_id"] == media_group_id]
async def daily_summary(client: Client):
@@ -310,7 +308,7 @@ async def daily_summary(client: Client):
logger.info(f"Summary chat {source_chat_id}, send results to {target_chat_id}")
# fake message
message = Message(
- id=0,
+ id=rand_number(),
chat=Chat(id=target_chat_id),
text=f"/summary #{duration}h cid={to_int(source_chat_id)}", # type: ignore
)
src/llm/utils.py
@@ -15,7 +15,7 @@ from markitdown import MarkItDown
from pyrogram.parser.markdown import BLOCKQUOTE_EXPANDABLE_DELIM, BLOCKQUOTE_EXPANDABLE_END_DELIM
from config import DOWNLOAD_DIR, GEMINI, GPT, PREFIX, cache
-from utils import nowdt, number_to_emoji, read_text, remove_consecutive_newlines, remove_dash, remove_pound, zhcn
+from utils import nowdt, number_to_emoji, read_text, remove_consecutive_newlines, remove_dash, remove_pound, strings_list, zhcn
BOT_TIPS = "（回复以继续）" # noqa: RUF001
REASONING_BEGIN = "🤔" # use emoji to separate model reasoning and content
@@ -178,7 +178,7 @@ def image_emoji(capability: bool) -> str: # noqa: FBT001
def clean_cmd_prefix(text: str) -> str:
- for prefix in [x.strip() for x in PREFIX.GPT.split(",") if x.strip()] + [PREFIX.GENIMG]:
+ for prefix in [*strings_list(PREFIX.GPT), PREFIX.GENIMG]:
text = text.removeprefix(prefix).lstrip()
return text
@@ -186,13 +186,13 @@ def clean_cmd_prefix(text: str) -> str:
def clean_bot_tips(text: str) -> str:
if not text:
return ""
- return re.sub(rf"(.*?){BOT_TIPS}", "", text, flags=re.DOTALL).strip()
+ return re.sub(rf"^🤖(.*?){BOT_TIPS}", "", text, flags=re.DOTALL).strip()
def clean_reasoning(text: str) -> str:
if not text:
return ""
- text = re.sub(rf"{REASONING_BEGIN}(.*?){REASONING_END}", "", text, flags=re.DOTALL).strip()
+ text = re.sub(rf"^{REASONING_BEGIN}(.*?){REASONING_END}", "", text.strip(), flags=re.DOTALL).strip()
text = text.removeprefix(BLOCKQUOTE_EXPANDABLE_DELIM).lstrip()
return text.removeprefix(BLOCKQUOTE_EXPANDABLE_END_DELIM).lstrip()
src/messages/chat_history.py
@@ -10,7 +10,7 @@ from config import MAX_MESSAGE_RETRIEVED, TZ
from messages.parser import parse_msg
-async def get_parsed_chat_history(
+async def get_history_info_list(
client: Client,
chat_id: int | str,
offset_id: int = 0,
@@ -43,7 +43,7 @@ async def get_parsed_chat_history(
break
if msg.empty:
break
- info = parse_msg(msg, silent=True)
+ info = parse_msg(msg, silent=True, use_cache=False)
if info["datetime"] < begin_time:
break
if info["datetime"] > end_time:
src/config.py
@@ -336,14 +336,8 @@ class GPT:
DOUBAO_BASE_URL = os.getenv("GPT_DOUBAO_BASE_URL", "https://ark.cn-beijing.volces.com/api/v3")
DOUBAO_ACCEPT_IMAGE = os.getenv("GPT_DOUBAO_ACCEPT_IMAGE", "1").lower() in ["1", "y", "yes", "t", "true", "on"]
- # /summary command
- SUMMARY_MODEL = os.getenv("GPT_SUMMARY_MODEL", "")
- SUMMARY_MODEL_NAME = os.getenv("GPT_SUMMARY_MODEL_NAME", "")
- SUMMARY_MODEL_MAX_OUTPUT_LENGTH = os.getenv("GPT_SUMMARY_MODEL_MAX_OUTPUT_LENGTH", "8192") # 8K
- SUMMARY_API_KEY = os.getenv("GPT_SUMMARY_API_KEY", "")
- SUMMARY_BASE_URL = os.getenv("GPT_SUMMARY_BASE_URL", "https://api.openai.com/v1")
- SUMMARY_TIMEOUT = os.getenv("GPT_SUMMARY_TIMEOUT", "600") # should be larger than default timeout
- SUMMARY_ACCEPT_IMAGE = os.getenv("GPT_SUMMARY_ACCEPT_IMAGE", "1").lower() in ["1", "y", "yes", "t", "true", "on"]
+ # AI summary (/summary)
+ SUMMARY_CMD = os.getenv("GPT_SUMMARY_CMD", "/gemini") # add this command prefix to call AI summary
# For tool_call. Some models doesn't support tool call, so we use this model to do the tool_call first.
# Then construct the new questions for the original model.