Commit 8ec8fbd
Changed files (9)
src/llm/contexts.py
@@ -58,7 +58,7 @@ async def single_gpt_context(client: Client, message: Message) -> dict:
}
"""
info = parse_msg(message, silent=True)
- role = "assistant" if f"{BOT_TIPS})" in info["text"] else "user"
+ role = "assistant" if BOT_TIPS in info["text"] else "user"
if info["mtype"] not in ["text", "photo", "voice", "video", "document"]:
return {}
@@ -118,7 +118,7 @@ async def single_gemini_context(client: Client, message: Message) -> dict:
}
"""
info = parse_msg(message, silent=True)
- role = "model" if f"{BOT_TIPS})" in info["text"] else "user"
+ role = "model" if BOT_TIPS in info["text"] else "user"
if info["mtype"] not in ["text", "photo", "voice", "video", "document"]:
return {}
extra_mime_types = ["application/pdf", "application/x-javascript"] # gemini has built-in support for these
src/llm/gemini.py
@@ -11,11 +11,11 @@ from google.genai.types import ContentUnionDict, GenerateContentConfig, GoogleSe
from loguru import logger
from PIL import Image
from pyrogram.client import Client
-from pyrogram.types import Message
+from pyrogram.types import Message, ReplyParameters
-from config import DOWNLOAD_DIR, GEMINI, PREFIX, TEXT_LENGTH
+from config import DOWNLOAD_DIR, GEMINI, GPT, PREFIX, TEXT_LENGTH
from llm.contexts import get_conversation_contexts
-from llm.utils import BOT_TIPS, beautify_llm_response, clean_gemini_sourcemarks, clean_prefix, clean_source_marks
+from llm.utils import BOT_TIPS, beautify_llm_response, clean_cmd_prefix, clean_gemini_sourcemarks, clean_source_marks
from messages.parser import parse_msg
from messages.progress import modify_progress
from messages.sender import send2tg
@@ -50,7 +50,7 @@ async def gemini_response(client: Client, message: Message, conversations: list[
response_modalities = ["TEXT", "IMAGE"] if modality == "image" else ["TEXT"]
tools = [Tool(google_search=GoogleSearch())] if modality == "text" else None
try:
- msg = f"🤖**{model_name}**: 思考中...\n👤**[{info['full_name']}](tg://user?id={info['uid']})**:“{clean_prefix(info['text'])}”"
+ msg = f"🤖**{model_name}**: 思考中...\n👤**[{info['full_name']}](tg://user?id={info['uid']})**: “{clean_cmd_prefix(info['text'])}”"
status_msg = (await send2tg(client, message, texts=msg, **kwargs))[0]
kwargs["progress"] = status_msg
contexts = await get_conversation_contexts(client, conversations, ctx_format="gemini")
@@ -78,31 +78,46 @@ async def gemini_stream(
retry: int = 0,
**kwargs,
):
- prefix = f"🤖**{model_name}**: ({BOT_TIPS})\n"
+ prefix = f"🤖**{model_name}**:{BOT_TIPS}\n"
+ answers = ""
try:
- status = kwargs.get("progress")
+ status: Message = kwargs.get("progress") # type: ignore
api_keys = [x.strip() for x in GEMINI.API_KEYS.split(",") if x.strip()]
if retry > len(api_keys) - 1:
return None
app = genai.Client(api_key=api_keys[retry], http_options=HttpOptions(base_url=GEMINI.BASR_URL, async_client_args={"proxy": GEMINI.PROXY}))
- answers = ""
+ runtime_texts = ""
async for chunk in await app.aio.models.generate_content_stream(**params):
resp = parse_response(chunk.model_dump())
answer = resp.get("texts", "")
+ runtime_texts += answer
answers += answer
- answers = beautify_llm_response(answers)
- if await count_without_entities(prefix + answers) <= TEXT_LENGTH:
- if len(answers.removeprefix(prefix)) > 10: # start response if answer is not empty
- await modify_progress(message=status, text=prefix + answers, detail_progress=True)
+ runtime_texts = beautify_llm_response(runtime_texts)
+ length = await count_without_entities(prefix + runtime_texts)
+ if length <= TEXT_LENGTH:
+ if len(runtime_texts.removeprefix(prefix)) > 10: # start response if answer is not empty
+ await modify_progress(message=status, text=prefix + runtime_texts, detail_progress=True)
else: # answers is too long, split it into multiple messages
- parts = await smart_split(prefix + answers)
+ parts = await smart_split(prefix + runtime_texts)
+ if len(parts) == 1:
+ continue
await modify_progress(message=status, text=blockquote(parts[0]), force_update=True) # force send the first part
- answers = parts[-1] # keep the last part
- status = await client.send_message(message.chat.id, answers) # the new message
+ runtime_texts = parts[-1] # keep the last part
+ status = await client.send_message(message.chat.id, text=prefix + runtime_texts, reply_parameters=ReplyParameters(message_id=status.id)) # the new message
# all chunks are processed
- answers = blockquote(beautify_llm_response(answers)) # blockquote AI response
- await modify_progress(message=status, text=prefix + answers, force_update=True)
+ if not answers.strip(): # empty response
+ return await gemini_stream(client, message, model_name, params, retry + 1, **kwargs) # type: ignore
+
+ if await count_without_entities(prefix + answers) <= TEXT_LENGTH: # short answer in single msg
+ if length > GPT.COLLAPSE_LENGTH: # collapse the response if the answer is too long
+ await modify_progress(message=status, text=f"{prefix}{blockquote(runtime_texts)}", force_update=True)
+ else:
+ await modify_progress(message=status, text=f"{prefix}{runtime_texts}", force_update=True)
+ elif length > GPT.COLLAPSE_LENGTH:
+ await modify_progress(message=status, text=prefix + blockquote(runtime_texts), force_update=True)
+ else:
+ await modify_progress(message=status, text=prefix + runtime_texts, force_update=True)
except Exception as e:
logger.error(e)
@@ -132,7 +147,7 @@ async def gemini_nonstream(
return None
app = genai.Client(api_key=api_keys[retry], http_options=HttpOptions(base_url=GEMINI.BASR_URL, async_client_args={"proxy": GEMINI.PROXY}))
response = await app.aio.models.generate_content(**params)
- prefix = f"🤖**{model_name}**: ({BOT_TIPS})\n"
+ prefix = f"🤖**{model_name}**:{BOT_TIPS}\n"
res = parse_response(response.model_dump(), prefix=prefix)
await send2tg(client, message, caption_above=True, **res, **kwargs)
await modify_progress(del_status=True, **kwargs)
@@ -148,6 +163,7 @@ async def gemini_nonstream(
def parse_response(data: dict, prefix: str = "") -> dict:
+ """Parse gemini response, includes texts, image and websearch."""
logger.trace(data)
parts = glom(data, "candidates.0.content.parts", default=[]) or []
gemini_logging(parts)
@@ -174,6 +190,7 @@ def parse_response(data: dict, prefix: str = "") -> dict:
def gemini_logging(contexts: list):
+ """Print logs of gemini contexts."""
msg = ""
with contextlib.suppress(Exception):
for item in contexts:
src/llm/gpt.py
@@ -13,7 +13,7 @@ from llm.models import get_context_type, get_gpt_config, get_model_id
from llm.response import send_to_gpt
from llm.response_stream import send_to_gpt_stream
from llm.tools import merge_tools_response
-from llm.utils import BOT_TIPS, clean_prefix, image_emoji, llm_cleanup_files
+from llm.utils import BOT_TIPS, clean_cmd_prefix, image_emoji, llm_cleanup_files
from messages.parser import parse_msg
from messages.progress import modify_progress
from messages.sender import send2tg
@@ -44,7 +44,11 @@ HELP = f"""🤖**GPT对话**
def is_gpt_conversation(message: Message) -> bool:
info = parse_msg(message)
- if info["is_bot"]: # do not process bot message
+ # to avoid potential infinitely loop,
+ # we do not respond to bot message & GPT responses.
+ if info["is_bot"]:
+ return False
+ if BOT_TIPS in info["text"]:
return False
if startswith_prefix(info["text"], prefix=[PREFIX.GPT, PREFIX.GENIMG, "/gpt", "/gemini", "/ds", "/qwen", "/doubao", "/grok"]):
return True
@@ -109,7 +113,7 @@ async def gpt_response(client: Client, message: Message, *, gpt_stream: bool = G
config["completions"]["messages"] = await get_conversation_contexts(client, conversations)
- msg = f"🤖**{config['friendly_name']}**: 思考中...\n👤**[{info['full_name']}](tg://user?id={info['uid']})**:“{clean_prefix(info['text'])}”"
+ msg = f"🤖**{config['friendly_name']}**: 思考中...\n👤**[{info['full_name']}](tg://user?id={info['uid']})**: “{clean_cmd_prefix(info['text'])}”"
status_msg = (await send2tg(client, message, texts=msg, **kwargs))[0]
kwargs["progress"] = status_msg
if context_type.get("error"):
@@ -118,7 +122,7 @@ async def gpt_response(client: Client, message: Message, *, gpt_stream: bool = G
config, response = await merge_tools_response(config, **kwargs)
# skip send a new request if tool_model is the same as the current model
if response and config["completions"]["model"] == GPT.TOOLS_MODEL and response.get("content"):
- texts = f"🤖**{config['friendly_name']}**: ({BOT_TIPS})\n\n{response['content']}"
+ texts = f"🤖**{config['friendly_name']}**:{BOT_TIPS}\n\n{response['content']}"
length = await count_without_entities(texts)
if length <= TEXT_LENGTH:
await modify_progress(message=status_msg, text=texts, force_update=True, **kwargs)
@@ -133,9 +137,9 @@ async def gpt_response(client: Client, message: Message, *, gpt_stream: bool = G
if content := response.get("content"):
if reasoning := response.get("reasoning"):
content = f"{reasoning}\n{content}"
- texts = f"🤖**{response['model']}**: ({BOT_TIPS})\n{content}"
+ texts = f"🤖**{response['model']}**:{BOT_TIPS}\n{content}"
else:
- texts = f"🤖**{response['model']}**: ({BOT_TIPS})\n\n{content}"
+ texts = f"🤖**{response['model']}**:{BOT_TIPS}\n\n{content}"
logger.debug(texts)
await send2tg(client, message, texts=texts, **kwargs)
await modify_progress(message=status_msg, del_status=True, **kwargs)
src/llm/response.py
@@ -10,7 +10,7 @@ from pyrogram.parser.markdown import BLOCKQUOTE_EXPANDABLE_DELIM, BLOCKQUOTE_EXP
from config import GPT
from llm.hooks import pre_hooks
-from llm.utils import add_search_results_to_response, beautify_llm_response, beautify_model_name, extract_reasoning
+from llm.utils import REASONING_BEGIN, REASONING_END, add_search_results_to_response, beautify_llm_response, beautify_model_name, extract_reasoning
from messages.progress import modify_progress
@@ -92,7 +92,7 @@ async def parse_response(config: dict, response: dict) -> dict[str, str]:
reasoning = glom(choice, Coalesce("message.reasoning_content", "message.reasoning"), default="") or ""
if reasoning and str(reasoning) != "None": # add expandable block quotation mark for reasoning
# if change this line, remember to remove the reasoning from contexts (`llm/contexts.py`)
- reasoning = f"{BLOCKQUOTE_EXPANDABLE_DELIM}🤔{reasoning.strip()}💡\n{BLOCKQUOTE_EXPANDABLE_END_DELIM}"
+ reasoning = f"{BLOCKQUOTE_EXPANDABLE_DELIM}{REASONING_BEGIN}{reasoning.strip()}{REASONING_END}\n{BLOCKQUOTE_EXPANDABLE_END_DELIM}"
primary_model = glom(config, "completions.model", default="") or ""
used_model = glom(response, "model", default="") or ""
src/llm/response_stream.py
@@ -9,13 +9,13 @@ from loguru import logger
from openai import AsyncOpenAI
from pyrogram.client import Client
from pyrogram.parser.markdown import BLOCKQUOTE_EXPANDABLE_DELIM, BLOCKQUOTE_EXPANDABLE_END_DELIM
-from pyrogram.types import Message
+from pyrogram.types import Message, ReplyParameters
from config import GPT, TEXT_LENGTH
from llm.hooks import pre_hooks
-from llm.utils import BOT_TIPS, add_search_results_to_response, beautify_llm_response
+from llm.utils import BOT_TIPS, REASONING_BEGIN, REASONING_END, add_search_results_to_response, beautify_llm_response, split_reasoning
from messages.progress import modify_progress
-from messages.utils import count_without_entities, smart_split
+from messages.utils import blockquote, count_without_entities, smart_split
async def send_to_gpt_stream(client: Client, status: Message, config: dict, retry: int = 0, **kwargs) -> dict:
@@ -25,13 +25,13 @@ async def send_to_gpt_stream(client: Client, status: Message, config: dict, retr
{"content": str, "reasoning": str, "model": str}
"""
# ruff: noqa: RUF001, RUF003
- prefix = f"🤖**{config['friendly_name']}**: ({BOT_TIPS})\n"
+ prefix = f"🤖**{config['friendly_name']}**:{BOT_TIPS}\n"
try:
pre_hooks(config["client"], config["completions"], message_info=kwargs.get("message_info"))
openai = AsyncOpenAI(**config["client"])
logger.trace(config)
answers = prefix
- sent_answers = []
+ all_answers = ""
is_reasoning = False
reasoning_in_response = None
gen = await openai.chat.completions.create(**config["completions"], stream=True)
@@ -50,12 +50,12 @@ async def send_to_gpt_stream(client: Client, status: Message, config: dict, retr
reasoning_in_response = True
if reasoning_content and not is_reasoning: # 首次收到推理内容
is_reasoning = True
- answers += f"{BLOCKQUOTE_EXPANDABLE_DELIM}🤔{reasoning_content.lstrip()}"
+ answers += f"{BLOCKQUOTE_EXPANDABLE_DELIM}{REASONING_BEGIN}{reasoning_content.lstrip()}"
elif reasoning_content and is_reasoning: # 收到推理内容且正在思考
answers += reasoning_content
elif reasoning_in_response is True and is_reasoning: # 收到回答, 关闭推理标志
is_reasoning = False
- answers = f"{answers.rstrip()}💡\n{BLOCKQUOTE_EXPANDABLE_END_DELIM}\n" + answer.lstrip()
+ answers = f"{answers.rstrip()}{REASONING_END}\n{BLOCKQUOTE_EXPANDABLE_END_DELIM}\n" + answer.lstrip()
else:
answers += answer
@@ -63,46 +63,49 @@ async def send_to_gpt_stream(client: Client, status: Message, config: dict, retr
# handle "<think>...</think>\n\n"
if answers.removeprefix(prefix).lstrip().startswith("<think>"):
is_reasoning = True
- answers = answers.replace("<think>", f"{BLOCKQUOTE_EXPANDABLE_DELIM}🤔")
+ answers = answers.replace("<think>", f"{BLOCKQUOTE_EXPANDABLE_DELIM}{REASONING_BEGIN}")
if "</think>" in answers:
is_reasoning = False
- answers = re.sub(r"</think>\s*", f"💡\n{BLOCKQUOTE_EXPANDABLE_END_DELIM}", answers, count=1)
-
- # handle ">Reasoning...Reasoned(.*?)seconds"
- if re.search(r"^>?\s*Reasoning", answers.removeprefix(prefix).lstrip(), re.DOTALL):
- is_reasoning = True
- answers = re.sub(r">?\s*Reasoning\s*", f"{BLOCKQUOTE_EXPANDABLE_DELIM}🤔", answers, count=1, flags=re.DOTALL)
- if re.search(r"🤔(.*?)Reasoned(.*?)seconds", answers.removeprefix(prefix).lstrip(), re.DOTALL):
- is_reasoning = False
- answers = re.sub(r"Reasoned(.*?)seconds\s*", f"💡\n{BLOCKQUOTE_EXPANDABLE_END_DELIM}", answers, count=1, flags=re.DOTALL)
-
- # handle ">正在推理...，持续(.*?)秒"
- if re.search(r"^>?(正在)?推理", answers.removeprefix(prefix).lstrip(), re.DOTALL):
- is_reasoning = True
- answers = re.sub(r">?(正在)?推理\s*", f"{BLOCKQUOTE_EXPANDABLE_DELIM}🤔", answers, count=1, flags=re.DOTALL)
- if re.search(r"🤔(.*?)，持续(.*?)秒", answers.removeprefix(prefix).lstrip(), re.DOTALL):
- is_reasoning = False
- answers = re.sub(r"，持续(.*?)秒\s*", f"💡\n{BLOCKQUOTE_EXPANDABLE_END_DELIM}", answers, count=1, flags=re.DOTALL)
+ answers = re.sub(r"</think>\s*", f"{REASONING_END}\n{BLOCKQUOTE_EXPANDABLE_END_DELIM}", answers, count=1)
answers = beautify_llm_response(answers)
- if await count_without_entities(answers) <= TEXT_LENGTH:
- if len(answers.removeprefix(prefix)) > 3: # start response if answer is not empty
+ if await count_without_entities(answers) <= TEXT_LENGTH - 10: # leave some flexibility
+ if len(answers.removeprefix(prefix)) > 10: # start response if answer is not empty
await modify_progress(message=status, text=answers, detail_progress=True)
else: # answers is too long, split it into multiple messages
parts = await smart_split(answers)
- await modify_progress(message=status, text=parts[0], force_update=True) # force send the first part
- sent_answers.append(parts[0])
+ if len(parts) == 1:
+ continue
+ if parts[0].startswith(prefix): # remove prefix from the first part
+ reasoning_resp, response = split_reasoning(parts[0])
+ content = reasoning_resp + "\n\n" + response
+ await modify_progress(message=status, text=f"{prefix}{blockquote(content.strip())}", force_update=True)
+ else:
+ await modify_progress(message=status, text=blockquote(parts[0]), force_update=True)
+ all_answers += parts[0]
answers = parts[-1] # keep the last part
if is_reasoning:
answers = f"{BLOCKQUOTE_EXPANDABLE_DELIM}{answers.lstrip()}"
- status = await client.send_message(status.chat.id, answers)
-
- sent_answers.append(answers)
- answers = add_search_results_to_response(config.get("search_results", []), "".join(sent_answers))
- answers = (await smart_split(answers))[-1]
- # Finally, force update the message
- await modify_progress(message=status, text=beautify_llm_response(answers), force_update=True)
+ status = await client.send_message(status.chat.id, text=prefix + answers, reply_parameters=ReplyParameters(message_id=status.id))
+ # all chunks are processed
+ all_answers += answers
+ all_answers = add_search_results_to_response(config.get("search_results", []), all_answers)
+ length = await count_without_entities(answers)
+ answers = (await smart_split(all_answers))[-1] # the last part (or the only one part)
+ reasoning_resp, response = split_reasoning(answers)
+ if answers.startswith(prefix):
+ if length > GPT.COLLAPSE_LENGTH:
+ content = reasoning_resp + "\n\n" + response
+ await modify_progress(message=status, text=f"{prefix}{blockquote(content.strip())}", force_update=True)
+ elif reasoning_resp:
+ await modify_progress(message=status, text=f"{prefix}{blockquote(reasoning_resp)}\n{response}", force_update=True)
+ else:
+ await modify_progress(message=status, text=f"{prefix}{response}", force_update=True)
+ elif length > GPT.COLLAPSE_LENGTH:
+ await modify_progress(message=status, text=prefix + blockquote(response), force_update=True)
+ else:
+ await modify_progress(message=status, text=prefix + response, force_update=True)
except Exception as e:
error = f"🤖{config['friendly_name']}请求失败, 重试次数: {retry + 1}/{GPT.MAX_RETRY + 1}\n{e}"
logger.error(error)
src/llm/utils.py
@@ -7,11 +7,14 @@ from typing import BinaryIO
import tiktoken
from loguru import logger
from markitdown import MarkItDown
+from pyrogram.parser.markdown import BLOCKQUOTE_EXPANDABLE_DELIM, BLOCKQUOTE_EXPANDABLE_END_DELIM
from config import DOWNLOAD_DIR, GPT, PREFIX
from utils import number_to_emoji, remove_consecutive_newlines, remove_dash, remove_pound, zhcn
-BOT_TIPS = "回复以继续"
+BOT_TIPS = "（回复以继续）" # noqa: RUF001
+REASONING_BEGIN = "🤔" # use emoji to separate model reasoning and content
+REASONING_END = "💡"
def llm_cleanup_files(messages: list[dict]):
@@ -133,18 +136,18 @@ def extract_reasoning(text: str) -> tuple[str, str]:
{content}"
"""
reasoning = ""
- if matched := re.search(r"^<think>(.*?)</think>", text.lstrip(), re.DOTALL):
+ if matched := re.search(r"^<think>(.*?)</think>", text.lstrip(), flags=re.DOTALL):
reasoning = matched.group(1)
text = re.sub(r"<think>(.*?)</think>", "", text, count=1, flags=re.DOTALL) # remove <think>...</think>
- if matched := re.search(r"^<thinking>(.*?)</thinking>", text.lstrip(), re.DOTALL):
+ if matched := re.search(r"^<thinking>(.*?)</thinking>", text.lstrip(), flags=re.DOTALL):
reasoning = matched.group(1)
text = re.sub(r"<thinking>(.*?)</thinking>", "", text, count=1, flags=re.DOTALL)
# Reverse engineered Web API
- if matched := re.search(r"^>?(正在)?推理(.*?)(，持续.*?)秒\n\n(.*)", text.lstrip(), re.DOTALL): # noqa: RUF001
+ if matched := re.search(r"^>?(正在)?推理(.*?)(，持续.*?)秒\n\n(.*)", text.lstrip(), flags=re.DOTALL): # noqa: RUF001
reasoning = matched.group(2)
text = matched.group(4)
- if matched := re.search(r"^>?\s?Reasoning(.*?)Reasoned(.*?)seconds\n\n(.*)", text.lstrip(), re.DOTALL):
+ if matched := re.search(r"^>?\s?Reasoning(.*?)Reasoned(.*?)seconds\n\n(.*)", text.lstrip(), flags=re.DOTALL):
reasoning = matched.group(1)
text = matched.group(3)
@@ -169,21 +172,31 @@ def image_emoji(capability: bool) -> str: # noqa: FBT001
return "(🏞)" if capability else ""
-def clean_prefix(text: str) -> str:
+def clean_cmd_prefix(text: str) -> str:
for prefix in [PREFIX.GPT, PREFIX.GENIMG, "/gpt", "/gemini", "/ds", "/qwen", "/grok", "/doubao"]:
text = text.removeprefix(prefix).lstrip()
return text
-def clean_response(text: str) -> str:
- """Remove bot prefix and reasoning content."""
+def clean_bot_tips(text: str) -> str:
if not text:
return ""
- text = clean_prefix(text)
- # remove bot tips
- text = re.sub(rf"(.*?){BOT_TIPS}\)", "", text, flags=re.DOTALL).strip()
- # remove reasoning
- return re.sub(r"^🤔(.*?)💡", "", 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 = text.removeprefix(BLOCKQUOTE_EXPANDABLE_DELIM).lstrip()
+ return text.removeprefix(BLOCKQUOTE_EXPANDABLE_END_DELIM).lstrip()
+
+
+def clean_response(text: str) -> str:
+ """Remove bot prefix and reasoning content."""
+ text = clean_cmd_prefix(text)
+ text = clean_bot_tips(text)
+ return clean_reasoning(text)
def clean_gemini_sourcemarks(contexts: list[dict]) -> None:
@@ -207,3 +220,20 @@ def convert_md(path: str | Path | BinaryIO) -> str:
logger.error(f"Failed to convert to markdown: {e}")
return ""
return result.text_content
+
+
+def split_reasoning(text: str) -> tuple[str, str]:
+ """Split reasoning from text.
+
+ Args:
+ text: LLM response
+ Returns:
+ (reasoning, content)
+ """
+ text = clean_cmd_prefix(text)
+ text = clean_bot_tips(text)
+ content = clean_reasoning(text)
+ reasoning = ""
+ if matched := re.search(rf"{REASONING_BEGIN}(.*?){REASONING_END}", text, flags=re.DOTALL):
+ reasoning = REASONING_BEGIN + matched.group(1) + REASONING_END
+ return reasoning.strip(), content.strip()
src/messages/progress.py
@@ -9,6 +9,7 @@ from pyrogram.errors import FloodWait, MessageNotModified
from pyrogram.types import Message
from config import TEXT_LENGTH, cache
+from messages.utils import count_without_entities, smart_split
async def modify_progress(
@@ -57,7 +58,9 @@ async def modify_progress(
if not detail_progress:
return
logger.trace(f"Progress: {text!r}")
- await message.edit_text(text[:TEXT_LENGTH])
+ if len(text) > TEXT_LENGTH and await count_without_entities(text) > TEXT_LENGTH:
+ text = (await smart_split(text))[0]
+ await message.edit_text(text)
cache.set("modify_progress", "1", ttl=ttl)
except FloodWait as e:
logger.warning(e)
src/messages/utils.py
@@ -130,7 +130,7 @@ async def smart_split(text: str, chars_per_string: int = TEXT_LENGTH, mode: Pars
def blockquote(s: str) -> str:
"""Block quote texts."""
- return BLOCKQUOTE_EXPANDABLE_DELIM + s + BLOCKQUOTE_EXPANDABLE_END_DELIM
+ return BLOCKQUOTE_EXPANDABLE_DELIM + s + "\n" + BLOCKQUOTE_EXPANDABLE_END_DELIM
def warp_comments(texts: str) -> str:
src/config.py
@@ -166,6 +166,7 @@ class GPT: # see `llm/README.md`
TOKEN_ENCODING = os.getenv("GPT_TOKEN_ENCODING", "o200k_base") # https://github.com/openai/tiktoken/blob/main/tiktoken/model.py
MAX_RETRY = int(os.getenv("GPT_MAX_RETRY", "2"))
HELICONE_API_KEY = os.getenv("HELICONE_API_KEY", "")
+ COLLAPSE_LENGTH = int(os.getenv("GEMINI_COLLAPSE_LENGTH", "500")) # Collapse the response if the length is greater than this value
# default command (/ai).
# set a string contains "gemini" to switch to gemini (see class GEMINI below for details)