generate_answer 'NoneType' object has no attribute 'encode'

[ahoepf]

Hello everyone, I am encountering the following error "user_text Test
generate_answer 'NoneType' object has no attribute 'encode'
cannot unpack non-iterable NoneType object
Exception in thread Thread-6 (thread_get_message):
Traceback (most recent call last):
File "/home/alberth/miniconda3/envs/textgen/lib/python3.10/threading.py", line 1016, in _bootstrap_inner
self.run()
File "/home/alberth/miniconda3/envs/textgen/lib/python3.10/threading.py", line 953, in run
self._target(*self._args, **self._kwargs)
File "/home/alberth/deeplearning/textgen/text-generation-webui/extensions/telegram_bot/TelegramBotWrapper.py", line 765, in thread_get_message
raise e
File "/home/alberth/deeplearning/textgen/text-generation-webui/extensions/telegram_bot/TelegramBotWrapper.py", line 746, in thread_get_message
answer, system_message = self.generate_answer(
TypeError: cannot unpack non-iterable NoneType object
" and unfortunately, I am unable to progress. I have configured everything according to the instructions. I am using the latest checked out version of text-generation-webui.

[innightwolfsleep]

Hi! I just made clear install of oobaboba and telegram_bot but I can't repeat your bug.

I am using llama7b4bit model and flags:
"--chat --gpu-memory 6 --cpu-memory 6 --model llama7b4bit --sdp-attention --api --extensions telegram_bot"

Can you share details about you command flags, prompt and config? I see you are not alone with this bug #119

[ahoepf]

Hi, thank you for response, i'm using this flags:
"--api --extensions superbooga silero_tts whisper_stt telegram_bot"
and when i tried your flags, no errors happened X-)

[sakhavhyand]

I've got the same symptoms today when playing with a new model.
The model auto-load with AutoGPTQ : no problem, I load it with ExLlama_HF : problem appear, load it again with ExLlama : problem disappear.
During all of this testing, everything was working fine through the WebUI.

It's strange because I have other models loaded via ExLlama_HF and they work well in the telegram integration.
The model causing problem in my case is TheBloke/MythoMax-L2-13B-GPTQ