[DredgenSun]

So ive been looking into building my own offline chat bot, but rather than be informative like ChatGPT, I want to just fill it with stuff from my own imagination, so it becomes a bot that's literally a clone. I've been building LoRA's and such for image diffusing, and wouldn't mind getting into text manipulation and chatbots, any ideas on good resources to look into? Should Breach have it's own LLM?

[HassaMassa]

Breached absolutely should not. It'll be filled with complete shit and the bias towards non-reality will be through the roof.

Building yourself something from a pre-existing LLM module would be a good step, I've been wanting for for a while but never get around to it.

https://huggingface.co/spaces/HuggingFac...eaderboard

I often check here to see what the best performing ones are.

[DredgenSun]

Breached absolutely should not. It'll be filled with complete shit and the bias towards non-reality will be through the roof.

Building yourself something from a pre-existing LLM module would be a good step, I've been wanting for for a while but never get around to it.

https://huggingface.co/spaces/HuggingFac...eaderboard

I often check here to see what the best performing ones are.

Thank you Hassa! (Great hearing from you again!)

As much as a Breach LLM might be reality bending.... huh... I guess that's a point in itself to create it, see where the HELL it ends up.

Essentially I'm trying to create some measure of therapy, by putting myself into an LLM, and then talking to myself, like i want to be hyper-self-aware lol

[HassaMassa]

Breached absolutely should not. It'll be filled with complete shit and the bias towards non-reality will be through the roof.

Building yourself something from a pre-existing LLM module would be a good step, I've been wanting for for a while but never get around to it.

https://huggingface.co/spaces/HuggingFac...eaderboard

I often check here to see what the best performing ones are.

Thank you Hassa! (Great hearing from you again!)

As much as a Breach LLM might be reality bending.... huh... I guess that's a point in itself to create it, see where the HELL it ends up.

Essentially I'm trying to create some measure of therapy, by putting myself into an LLM, and then talking to myself, like i want to be hyper-self-aware lol

I guess the main issue is the training data itself. That's the hardest thing to compile. Pre-existing ones like llama will have a lot to work with already but you'll never be able to guarantee a lack of bias etc if you don't provide your own data.

[DredgenSun]

Breached absolutely should not. It'll be filled with complete shit and the bias towards non-reality will be through the roof.

Building yourself something from a pre-existing LLM module would be a good step, I've been wanting for for a while but never get around to it.

https://huggingface.co/spaces/HuggingFac...eaderboard

I often check here to see what the best performing ones are.

Thank you Hassa! (Great hearing from you again!)

As much as a Breach LLM might be reality bending.... huh... I guess that's a point in itself to create it, see where the HELL it ends up.

Essentially I'm trying to create some measure of therapy, by putting myself into an LLM, and then talking to myself, like i want to be hyper-self-aware lol

I guess the main issue is the training data itself. That's the hardest thing to compile. Pre-existing ones like llama will have a lot to work with already but you'll never be able to guarantee a lack of bias etc if you don't provide your own data.

Hmm, so i was thinking of loading a lot of Whatsapp messaging from my end, I've been training to a degree, and decided on a LoRa for my diffusing; rather than have an the checkpoin/model hold my face data, I put my face data into a LoRa, piped it into the workflow as an injection to interpret after the model has loaded all the prompts/reference imaging.

[HassaMassa]

Breached absolutely should not. It'll be filled with complete shit and the bias towards non-reality will be through the roof.

Building yourself something from a pre-existing LLM module would be a good step, I've been wanting for for a while but never get around to it.

https://huggingface.co/spaces/HuggingFac...eaderboard

I often check here to see what the best performing ones are.

Thank you Hassa! (Great hearing from you again!)

As much as a Breach LLM might be reality bending.... huh... I guess that's a point in itself to create it, see where the HELL it ends up.

Essentially I'm trying to create some measure of therapy, by putting myself into an LLM, and then talking to myself, like i want to be hyper-self-aware lol

I guess the main issue is the training data itself. That's the hardest thing to compile. Pre-existing ones like llama will have a lot to work with already but you'll never be able to guarantee a lack of bias etc if you don't provide your own data.

Hmm, so i was thinking of loading a lot of Whatsapp messaging from my end, I've been training to a degree, and decided on a LoRa for my diffusing; rather than have an the checkpoin/model hold my face data, I put my face data into a LoRa, piped it into the workflow as an injection to interpret after the model has loaded all the prompts/reference imaging.

I don't know shit about image AI I'm afraid but I'm not convinced it's the same process. You'd need a mega load of whatsapp data and context outside of that though, or presumably it has no way of taking any of the whattsapp messages into context. 

If you wanted it to analyse that is, you could just include the whatsapp messages and the LLM could be used as a glorified file / text parser of sorts.

[DredgenSun]

Breached absolutely should not. It'll be filled with complete shit and the bias towards non-reality will be through the roof.

Building yourself something from a pre-existing LLM module would be a good step, I've been wanting for for a while but never get around to it.

https://huggingface.co/spaces/HuggingFac...eaderboard

I often check here to see what the best performing ones are.

Thank you Hassa! (Great hearing from you again!)

As much as a Breach LLM might be reality bending.... huh... I guess that's a point in itself to create it, see where the HELL it ends up.

Essentially I'm trying to create some measure of therapy, by putting myself into an LLM, and then talking to myself, like i want to be hyper-self-aware lol

I guess the main issue is the training data itself. That's the hardest thing to compile. Pre-existing ones like llama will have a lot to work with already but you'll never be able to guarantee a lack of bias etc if you don't provide your own data.

Hmm, so i was thinking of loading a lot of Whatsapp messaging from my end, I've been training to a degree, and decided on a LoRa for my diffusing; rather than have an the checkpoin/model hold my face data, I put my face data into a LoRa, piped it into the workflow as an injection to interpret after the model has loaded all the prompts/reference imaging.

I don't know shit about image AI I'm afraid but I'm not convinced it's the same process. You'd need a mega load of whatsapp data and context outside of that though, or presumably it has no way of taking any of the whattsapp messages into context. 

If you wanted it to analyse that is, you could just include the whatsapp messages and the LLM could be used as a glorified file / text parser of sorts.

Hassa, you're a flippin' genius! I'll take on board what you said, run my messages through an LLM and see what comes from that.


Probably a diagnosis, knowing my luck