hello everyone who can provide a dataset that would be like a neurosama
in terms of fun
#need datasets
spice acorn
worldly forum
If you need it for an LLM, an open-source Neuro LLM is already coming soon
No need to train your own probably
spice acorn
vedal seemed to say that he would provide open source code when he got tired of it and it makes him a profit and it's his brainchild so I don't think
and if it comes out, then when about
worldly forum
vedal seemed to say that he would provide open source code when he got tired of ...
Not from Vedal, from the community
crystal bobcat
neurosame 
worldly forum
Not from Vedal, from the community
The model is community-made
spice acorn
so the community recreated the behavior?
crystal bobcat
Still not quite released though
i think there's some sort of misunderstanding
worldly forum
But close
worldly forum
`i think there's some sort of misunderstanding`
I'm pretty sure open-Neuro-13B is not released yet
spice acorn
and will there be some kind of application for management and communication, for example, like Vedal on discord and avatar?
crystal bobcat
I'm pretty sure open-Neuro-13B is not released yet
no i
nevermind
worldly forum
and will there be some kind of application for management and communication, for...
You can use an LLM loading application like text generation webui
spice acorn
so they don't have the same functionality as vedal?
worldly forum
Well, it can be programmed in with more stuff by interfacing with the webui's API, I have a Discord bot that does so
Probably not exactly what you want, but you can use the interface code as a base
spice acorn
and with the help of this bot, it will be possible to communicate with her through a discord like vedal
I would just like a copy of neurosoma with all the functionality
worldly forum
Trough Disocrd text chat at least, you'd need a STT and TTS bot for the voice part
crystal bobcat
I would just like a copy of neurosoma with all the functionality
not likely
but you can try for something similar
worldly forum
Yeah, it is not an exact replica, but it does decently well when ran correctly
spice acorn
I mean, communicating with her by voice and receiving a message from her by voice, too.
crystal bobcat
so if that was your question: Neuro is closed-source
spice acorn
in general, it will be difficult because I do not know English enough to talk to someone, I do not know if it will be possible to adapt it into Russian
worldly forum
I mean, communicating with her by voice and receiving a message from her by voic...
That would need a bot to talk with the other bot that takes input from a VC, STTs it to text, gives that to the other bot, then take the output, feed that into a TTS and then output the result in the VC
worldly forum
in general, it will be difficult because I do not know English enough to talk to...
Then just add a translation step
It can be done in the bot that interfaces between the VC and the bot
spice acorn
That would need a bot to talk with the other bot that takes input from a VC, STT...
this is hard to realize if i so so in python
worldly forum
this is hard to realize if i so so in python
I use Kotlin for Discord bot stuff, but there are already solutions for TTS and STT in existence you can hook into, you just need to look for them. For TTS, a solution that could work is coqui TTS
You just need something that has an API
spice acorn
and will it be difficult to realize that it will be possible to interrupt her when she speaks and also so that she can do something herself
spice acorn
You just need something that has an API
in what sense
worldly forum
in what sense
An API is usually something you can send a request into that will return some kind of data you want, in this case you want one API that returns the text from audio and one that returns the audio from text
worldly forum
You just send http requests into some documented API, that's how the LLM interface bot works
I kinda have an idea of how it would look in Kotlin
It would definitely include using Kord for Discord stuff, okhttp for http requests and probably kotlinx.serialization.json for formatting
Pretty simple stuff
Kotlin is really easy to learn if that's the main concern
spice acorn
by the way, regarding translation, if you translate a sentence from one language to another and back twice, almost the whole meaning is lost
worldly forum
You would only be translating each sentence once, Russian to English to feed to the LLM and then English to Russian to output
The sentence changes in the middle
I assume you meant to put that into a translator
spice acorn
yes
winter cypressBOT
You have unlocked new role
spice acorn
for example, I will translate one sentence twice "I assume you meant to put that into a translator" - "I suppose you were going to put this into the translator"
worldly forum
The bot would only need to translate one way, since the bot generates its own sentence
spice acorn
Oh yeah
worldly forum
One translation to English on your input and one translation to Russian on the output
spice acorn
yes, then the meaning will remain in the majority
worldly forum
Yeah, that would most likely work
You just need a library or API that can translate
spice acorn
how to make it so that sentences reach the model in chunks so that it can be interrupted
boreal narwhal
The model is community-made
Is there a repo or thread for this? It's nifty, first i heard of it, hadn't had luck searching discord for it.
spice acorn
I'm pretty sure open-Neuro-13B is not released yet
I tried to find it, but I didn't find anything related to neurosama
worldly forum
As I said, it's not released yet, but it'll probably be soon
spice acorn
But I didn't even find open-Neuro-12B
worldly forum
how to make it so that sentences reach the model in chunks so that it can be int...
That might need you to split it into multiple API calls, which would take rewriting the entire interface to work differently and make it require a lot more processing power
worldly forum
But I didn't even find open-Neuro-12B
The number before B is the number of LLM parameters, the Neuro model is a 13 billion parameters model
worldly forum
Is there a repo or thread for this? It's nifty, first i heard of it, hadn't had ...
I can get you a preview of it, you can DM me if you want that. The model is not released yet, but I assume it will be some time soon
spice acorn
I realized if you translate very quickly, then the voice is translated into the text slowly anyway, so it will recognize that you started talking and therefore it can be interrupted or recognize that the translation of something has begun
worldly forum
That is also a way to do it, although not as flexible
spice acorn
in what way is it not so flexible
worldly forum
It can't be stopped mid-generation, but if you don't need that, then it should be fine
spice acorn
is it possible to explain in more understandable words, it's just that the translator translates it very old
worldly forum
If you want to stop the bot while the bot is speaking, you can't do that as eaily without breaking the chatlogs with that method
spice acorn
and why not just give him a stop command, for example, at which the response generation stops and he finishes what's left when you start talking
or something like that
worldly forum
The generation being able to be stopped would require the LLM request to be split into multiple requests, which would slow things down greatly and make translation worse
spice acorn
well, then just stop the translation of text into audio or just interrupt it when, for example, she reads to the end of a sentence
worldly forum
The translation I'm referring to is the English to Russian step
That one basically needs the entire output for the best results
spice acorn
or you can ask what kind of speech recognizer you can use to make it respond quickly and consume little
worldly forum
The speech recognizer is not the issue with stopping midway, the issue is how we need to handle LLM requests and chatlogs
spice acorn
and also speech recognizers if you speak Russian and then you say some English term, it is poorly perceived
no, I just don't know which speech recognizer to use for the code, I tried to use one, but it worked so that you speak and it translates it into text in 4-5 seconds in the fastest case
worldly forum
Speech conversion requires a really large amount of processing power, Vedal definitly uses cloud compute for that too
spice acorn
No one says I want to process it on pu recommend a library that is fast for speech recognition on the cloud or on the internet
worldly forum
Then maybe the server you connected to was not fast enough or you had a lot of latency
spice acorn
Well, what library can you recommend?
modern lotus
If you need it for an LLM, an open-source Neuro LLM is already coming soon
are you talking about...
worldly forum
Well, what library can you recommend?
I don't know of any good library, but I recommend you look for something that works locally to compare with a cloud-based solution, in case the cloud is slower. You will need to figure things out
worldly forum
are you talking about...
Maybe
spice acorn
by the way, how did Vedal give ai a name so that she would respond to it and give her a character
modern lotus
by the way, how did Vedal give ai a name so that she would respond to it and giv...
fine-tuning
worldly forum
No, it's a prompting trick, I know from building an LLM interface bot
You just end the usermade prompt with "$charname:" and then the LLM will continue as if speaking as that character
That is of course all internal, so you don't notice it
sweet token
no one will just give you the secret sauce, you need to figure it out yourself.
worldly forum
Well, figuring out how stuff works is quite fun