Handle responses longer than telegram message limit

[AlexHTW]

Hey, I encountered the telegram error Message_too_long when transcribing a 6 minute audio file.
Any chance to split responses longer than the limit (I think 4096 characters) into multiple messages?
Might also apply to chatGPT responses although I have not managed to get such a long response yet but I think theoretically it could be possible (4096 tokens > 4096 characters).

Thank you and the other contributors for all your hard work!

[n3d1117]

Hi @AlexHTW, is this a Telegram or a ChatGPT limitation?

I don't have much time, but I added to the README in case someone wants to give it a try!

[AlexHTW]

Hey, it's a Telegram limitation. We would just need to split the response into multiple messages if response is larger than the Telegram limit. Thanks for adding it, maybe I can figure it out and contribute.

[n3d1117]

@AlexHTW this should be fixed with #76. Can you confirm this?

[AlexHTW]

Hey @n3d1117, thanks.
Yeah I managed to get split messages for a 30 minute voice message and also from prompting for very long answers. So I would say everything works as intended.

I tested with 50-60 minute podcast mp3s and couldn't get a transcript, unfortunately also no log message. Maybe it's an API limit or something about the files itself. Don't think it has to do with this PR though, so I would say ship it 🚀

[AlexHTW]

I noticed I don't receive a "new transcribe request received" info log when sending the bot podcast mp3s.
On the main branch I get the info log.
53min example file: https://content.blubrry.com/takeituneasy/lex_ai_noam_chomsky_2.mp3

[n3d1117]

@AlexHTW you're right. Bots can download files of up to 20MB in size (source) and the .mp3 you linked is 38,1MB.

There was actually an exception raised in the context.bot.get_file which wasn't handled. Should be improved now with f7bb416.

BTW I tried compressing your mp3 using an online tool, the download then worked but I got:

openai.error.APIError: Maximum content size limit (26214400) exceeded (50819087 bytes read) {
  "error": {
    "message": "Maximum content size limit (26214400) exceeded (50819087 bytes read)",
    "type": "server_error",
    "param": null,
    "code": null
  }
} 413

suggesting there's a limit (~26MB?) for Whisper requests too. I think you're better off splitting your audio into multiple files

[AlexHTW]

Thanks for the investigation! I was not aware of those limits.

I looked up the limitations in the Whisper docs, you are correct:

File uploads are currently limited to 25 MB and the following input file types are supported: mp3, mp4, mpeg, mpga, m4a, wav, and webm

The user feedback for files too large is important, thanks. I would suggest also telling the user the actual limit of 20MB when the file is too large.

[n3d1117]

Done! Merging now, thanks for testing

[AlexHTW]

Hey, it looks like the streaming update broke the handling for long messages in chat responses.
After the telegram message limit is reached I get another message "Failed to get response: Message_too_long".

chatgpt-telegram-bot_1  | Traceback (most recent call last):
chatgpt-telegram-bot_1  |   File "/app/bot/telegram_bot.py", line 370, in prompt
chatgpt-telegram-bot_1  |     await context.bot.edit_message_text(content, chat_id=sent_message.chat_id,
chatgpt-telegram-bot_1  |   File "/usr/local/lib/python3.9/site-packages/telegram/ext/_extbot.py", line 1453, in edit_message_text
chatgpt-telegram-bot_1  |     return await super().edit_message_text(
chatgpt-telegram-bot_1  |   File "/usr/local/lib/python3.9/site-packages/telegram/_bot.py", line 331, in decorator
chatgpt-telegram-bot_1  |     result = await func(*args, **kwargs)  # skipcq: PYL-E1102
chatgpt-telegram-bot_1  |   File "/usr/local/lib/python3.9/site-packages/telegram/_bot.py", line 3230, in edit_message_text
chatgpt-telegram-bot_1  |     return await self._send_message(
chatgpt-telegram-bot_1  |   File "/usr/local/lib/python3.9/site-packages/telegram/ext/_extbot.py", line 488, in _send_message
chatgpt-telegram-bot_1  |     result = await super()._send_message(
chatgpt-telegram-bot_1  |   File "/usr/local/lib/python3.9/site-packages/telegram/_bot.py", line 512, in _send_message
chatgpt-telegram-bot_1  |     result = await self._post(
chatgpt-telegram-bot_1  |   File "/usr/local/lib/python3.9/site-packages/telegram/_bot.py", line 419, in _post
chatgpt-telegram-bot_1  |     return await self._do_post(
chatgpt-telegram-bot_1  |   File "/usr/local/lib/python3.9/site-packages/telegram/ext/_extbot.py", line 306, in _do_post
chatgpt-telegram-bot_1  |     return await super()._do_post(
chatgpt-telegram-bot_1  |   File "/usr/local/lib/python3.9/site-packages/telegram/_bot.py", line 450, in _do_post
chatgpt-telegram-bot_1  |     return await request.post(
chatgpt-telegram-bot_1  |   File "/usr/local/lib/python3.9/site-packages/telegram/request/_baserequest.py", line 165, in post
chatgpt-telegram-bot_1  |     result = await self._request_wrapper(
chatgpt-telegram-bot_1  |   File "/usr/local/lib/python3.9/site-packages/telegram/request/_baserequest.py", line 328, in _request_wrapper
chatgpt-telegram-bot_1  |     raise BadRequest(message)
chatgpt-telegram-bot_1  | telegram.error.BadRequest: Message_too_long

example prompt: list all presidents of the United States with a short biography and list of their major achievements.

[n3d1117]

Good catch @AlexHTW, I got a fix and i'll push it tomorrow

[n3d1117]

Actually just pushed it now @AlexHTW, could you do a git pull and test it? bc6a4e4

[AlexHTW]

Awesome. Tested it, works great :) Thank you