LeroyDyer/Mixtral_AI_1.0_7b

LeroyDyer/Mixtral_AI_1.0_7b-GGUF

A VERY GOOD MODEL !

INSTRUCT

such as give me a entitylist of words pertaining to the various emotions ! or create a tokenizer in python/c#... . Very good instruct model using the instruct prompt template te model responds directly to the task requested.

CHAT BASED (turn based Prompting with history etc)

for day to day talking is varied as suprising. as well as taking on a reflective personality ie using the same speech paterns even . Formal/ causual / academic etc;

OpenFunctions

As a server the model serves functions and produces results to task creating functions in python or powershell / vbscript / bashcode ready to run

CODE MODEL

as a coder ! YES ! at last , responding to requests for code snipptets and explanitory code and concepts fully explained.

USES

The model can be used in various prompt structures which all produces results accordingly. it will perform all task in all modes with correct prompting. the model may respond angry or annoyed or even sarcastically! hence great for roleplay.

TRAINING:

This model has been trained on various data sets andis still undergoing fine tuning stages and difussion merges. as well as base model attention merges to reallign the model with basic original base model features.

Context Windows

This model was not extended to 128k ie with the rope embeddings and yarn.. so it is 32k which has been found to be a stable input window for 7b param models . there are other models which are being experimented with in parallel to this model which have the 128k context. but this also means that the dataset for training needs to contain samples of 128k context to realign the model to 128k context usage. extending context windows , requires some retraining. perhaps only for 2-3 epoch on a large dataset. in later training cycle a smaller dataset with higher epochs can also reInforce Basic methods and traits , such as formatted responses ; by giving examples of such formatted responses it enables for the model to create a higher level of output in its basic responses.

CURRENT_ CONCEPT

This model will also undergo a transformation enabling the ability to utilize the the image encoding in the tokenization process in later itterations also enabling for the model to communicate with image inputs as well as accept training of images . for image to text understanding . understanding the pipeline system and configuration of the models requirements that need to be satisfied to become the multi modal model. for image and text input and output. which also requires the clip tokenizer to be firstly injected into the model. (LLAVA MODELS & DIFFUSION MODELS)