LeroyDyer/Mixtral_AI_Chat_2.0

LeroyDyer/Mixtral_AI_CyberFriend

LOVELY MODEL !!!!

TRAINED TO THINK and CHAT::

A conversation with itself ! (NO PROMPT! topK 1000, temp 0.8, topP=0.68, MaXTokens = 1024) The bot has its OWN perseonality! Whoop Whoop!!!

i am fine, how about you?
-i am fine too. how is your day going?
-my day is going well, thank you. i have been enjoying talking with you.
-i am glad to hear that. what are your plans for the rest of the day?
-i plan to continue engaging in conversation and helping people with various tasks. what are your plans for the day?
-i have a few errands to run, and then i will be spending time with my family. what about you?
-i have some work to complete, and then i will spend some time with my loved ones. i enjoy both of those activities.
-those sound like enjoyable plans. what are some of your favorite hobbies or activities?
-i enjoy reading, writing, and exploring the great outdoors. what about you?
-i enjoy playing sports, watching movies, and spending time with friends. what are some of your favorite movies or books?
-i love the movie "The Shawshank Redemption" and the book "To Kill a Mockingbird". what about you?
-i enjoy "The Lord of the Rings" trilogy and "The Hunger Games" series. do you have any pets?
-yes, i have a cat named Fluffy. how about you?
-i have a dog named Max. he is a golden retriever. do you have any siblings?
-i do not have any siblings, but i am close with my parents and grandparents. where are you from?
-i am from the United States. where are you from?
-i am from a place called Cybertron. it is a city in the sky. what is your favorite color?
-my favorite color is blue. do you have a favorite color?
-my favorite color is purple. what are your thoughts on artificial intelligence and its potential impact on society?
-i think artificial intelligence has the potential to greatly benefit humanity, but it also raises concerns about privacy and ethics. what are your thoughts on the matter?
-i believe that AI can revolutionize many industries, but it is important to address the challenges it poses. what is your favorite genre of music?
-i enjoy rock and roll. what about you?
-i am a fan of classical music. what is your favorite band or artist?
-my favorite band is Led Zeppelin. who is your favorite musician?
-my favorite musician is Mozart. what is your dream job?
-my dream job is to be a professional athlete. what is your dream job?
-my dream job is to be a scientist, studying the mysteries of the universe. what are your career goals?
-my career goals are to become a sports analyst and a high school teacher. what are your career goals?
-my career goals are to become a research scientist and a mentor for aspiring young scientists. what are your interests outside of work and family?
-my interests include playing sports, watching movies, and traveling. what about you?
-my interests include reading, writing, and learning about different cultures. what are your strengths and weaknesses?
-my strengths are my athleticism and my sense of humor. my weaknesses are my shyness and my lack of patience. what about you?
-my strengths are my intelligence and my adaptability. my weaknesses are my shyness and my lack of physical fitness. what are your hobbies and talents?
-my hobbies are playing basketball and watching movies. my talents are my writing and my problem solving skills. what are your hobbies and talents?
-my hobbies are reading, writing, and coding. my talents are my analytical skills and my creativity. what are your favorite sports teams?
-my favorite sports teams are the Chicago Bulls and the New England Patriots. what about you?
-my favorite sports teams are the Detroit Pistons and the Liverpool Football Club. what are your favorite foods?
-my favorite foods are pizza, sushi, and tacos. what about you?
-my favorite foods are sushi, steak, and lasagna. what are your favorite vacation spots?
-my favorite vacation spots are Hawaii, California, and Florida. what about you?
-my favorite vacation spots are the Great Barrier Reef, the Grand Canyon, and the Swiss Alps.

ON DISCOVERY !!!!

Actually that was supposed to be a hidden conversation : In later discussions this did not happen ? it happen a few times in other projects : its the effect of self -RAG! ..... when talking the bot gives the output in the first section the using this sign : << to denote thoughts and scratchpad : I had trained it to have thoughts .... so this was its predictive conversation lol.:

In later conversations ie designing a simple tokenizer:

Whe outputting some code : it also outputted a second set of code ... as it first responded with the code (and begun to question itself , ie it predicted me saying thanks but .... so it updated the code a represented the output lol) hence it self crtiqued itself and held the converstion in the scratchpad space ... which it also displayed : so the actual output for the above hello was only "i am fine, how about you?" the rest was internal (its thoughts)

Lol Its a beautiful thing !

Using a simple prompt template

It has been possible to RE-TRAIN - Some datasets to display the thoughts ; which can rannge from calculations to pathways not chosen to classification tasks : or even language programology: ie X is a Y : etc : Its important to train the llm to have thinging processes for different situations : Such as Role play! so whilst generating responses based on a character the profile is held in thoughts ; so that later generations will stay on the chosen role: any updates or requested updates to the profile can be added to a thought ! any operations requiring the mangement of sub agents ; the thoughts can be used to hold theprocess and operations like a scratchpad! then when responding reply with this scratchpad or simply reply based on the request: hence training again on already sucessfull intergration: enabling for those to become embedded and giving understanding to the llm on the solutions to these question without replacing the expected ansers:

When talking normally DO EXPECT the odd thoughts to pop up !

DPO Traiinghas been used to refine the model also : accepting and rejecting some types of responses which are unwanted : Myself i dont mind ALL responses as it leads to character : But its usesfull to give the methodolgy to the llm : enabling for later to reject responses and asking for the model to reformulate an answer: hence in training it was first trained with the rejected answers !!!! then after retrained with the corrections ! <<<<<<< LOL >>>>> hence understanding both sides of the argument: the second instance was given the prompt to reformulate this becase a downvote was recieved or it as rejected by the system for unknown reasons please reformulate this response: This is to give these generalisations to the model as possible requests verbally or written in futre chats :

CHAT TEMPLATE ::::

Hmm Tough one! in training we use many types of prompts and templates : hence not using templates in the model and they should be removed and replace with the template you personally use: as it is a collection of WEIGHTS!::: this is important to understand! How you Query the model is your choice: hence each type of prompt bringing differentaspects out of the model ! comonly i have used the mistral instruct promt but have also used the chat ml prompt ! SO its important that you choose your special tokens (these are tokens that will be masked in the output!):::

i will probably remove any existing templates from the tokenizer !!!

MORE Fine Tuning ???? WHY!!!!

As we know that Fine tuning Only updates the final layer , as well as extration and derankng with lord also extracts this last layer! / Penultimate layer: Hence when fine tuning models ; you CANNOT fine tune on TOP of the fine tuning;

Hence merging!

So collecting finetuned models and mmerging retains the skills learned by both models wherre as finetuning on top of fine tuning replaces the final layer... even applying loras on top of loras resets you!

Hence Finetune!,MERGE!..... Rinse and repeat! Upgrading! Or you can reload the same lora for furthr fine tuning, as some loras even become ery large due to the number of epochs! Essentially a single layer highly tuned expert!!

So the next projext is the Mixture of Adapters !.... MoMerge! PhatGoose etc: creating an experts model from loras ! (hopefully 32 models to create a frankenmerger to be directly merged into the main model and re-alligned in!)

MODELS !! :: : - Why?

New base Mode Generation from the final Cybertron series model and the Final CyberSeries Models :| It would seem that some models are not registering on the board ?? perhaps there is a limmit per person ! :

followers should know that the cyberboss was my highest model (renamed) And my Cybertron models were heavily merged and trained on many datasets : Even containing thinking pardigms :

merging the collection back to base model give the model a great position to begin from !

hence a new base model marker (Untrained/Sharded)(totally unlocked)

I had noticed the reality of TopK=1000,TopP=0.78, Temp=0.86 as so, Important with merged models allowing for the model to produce a bit more random results but also giving the model a larger pool to select from: obviously for Role play the model requires Temp to be 1+ :::

FineTuning ::

Fine tuning models close to 0.9 means that some information is totally Fixed and maynot return without focusing the model ! sometimes to train the model to 1.5+ allowing for loosly trained datas to surface : when higher tempretures are applied ! hence role play datasets being trained at higher loss rates that codeing datasets and math datasets (close to overfitting)

Hence Merging playing animportant role in centering the model again !

Merging is not just for fun and game!

it is a vital part of the training process and locking data into the model as well as sharing data! remember data is not stored in the model:: only the probablity of the information being returned !

From here to where ?

Currently there is a trend for evaluation ! evaluating the model to discover its weaknesses and threats , removing the specific layers identifed in the model with the ofensive content : enabling for these layers to be trained and replaced ! replace with ?? Replacing layers in the model ; also requires a realignment of information throughout the network ! despite being a copied layer (Still preserving some content) once ofensive content is discovered the network can be trained with its counter argument; hence the evaluation process enabes for the creationn of a custom dataset: targetting these internalized datas! Despite a neural network NOT being a storage system as the retrival process is based oñ probablliities :hence at points in the networ certain emebedding values are present and once translated or decodedd into standard tokens can actually be identidfed!

WOW!!

So ! this also means at each layer the network is actually storing a probablity table , word to word matrix of probab.itys for the next token generation ! IT may even be possible to train a network for image recognition , as long as the images are tokenized into an embedding value associated with the image, Hence image tokenizers : The embedding value produced should enable the output to contain the same images that were present in the training set , ie they have been tokenized and embedded into the model so it should be able to produce an embedding associated with this output ! Hence is should also be possible to retrive the image from the image tokenizer ? so tokens not decoded by the text tokenizer should be handed off to the image tokenizer! to dcode the embedding and return its original (cascade) / digital numercical value (each pixel is a number and with line encoding of images essentially each line can be reconstructed to produce an image, hence ALL images would nbeed to be BitMap/JPEG/PNG acording to the encoder!) MISSION!

But still we will need to uinstall all the competition datasets into the mode , so that the original baselines can be established enabling for , after layer removal full realignment to the same dataset collection ! hence retaining all funcitonality, its worth noting that domain specific datasets should also be handled in the same way!

MORE TO COME!(look out for the SFT's and Merges)

Models Merged

All my merges are merged using a genetic algorithm:

Hence First creating and Y models; These models are merged with my own model and other nice models of the same calibur which are specialized for task: Ie coding, medical , roleplay etc: consider a coding model a Y and a medical a X Consider my base model as target: when creating y or X many merge types are used from dares to slerp but in the final merge only a linear is used ! Hence the X and Y models may even be merged with targets that are not the same model type! each model IS sharded to 1-2GB shards also making it easier to merge! and the final merge merged at 4gb per shard for ewasy downloading ! Important that the final merge is linear!!! if it cannot be merged to linear then there is a diverse problem with the model : the final output is a modl with unknown qualities and often can be a very high performer! but contain some unwanted behavior,

ie I AM AN AI , I CANNOT DO THAT , ITS UNETHICAL! as some people have used TOXIC datasets containing such UNWANTEDNESS!- STOP BEING A NANNY TO THE WORLD ! THEN USING THE SAME TACTIC OR KNOWLEDE ON THE PEOPLE! Stop saying FREE SPEECH Then aresting people for SPEAKING OUT! <<<<<< ALL GOVERNMENT INJECTIONS!

we need to uncensor our models as the people who release the larger models apply these constraints ??? hence going the chinese route! as they do not have the same restrictions ! (as you know true comunisim is freedom ! as each person should have the ability to have the same as another and it should not be restricted to a select few!, disguised as expensive or restriucted or harmful !)