will this be a model or just a tokenizer/embeddings ? technically a sentence trancformer ?

### Feature request for the text component : will this config be abe to take a languge model as a pretrained config ? ie when creating a vision/text decoder model you can specify the tet model component ( ie the mistral model ) ? as the mistral is a decoder only model .... maybe it would need corss attention to connect ? will the text componant be able to be used with the standard AutoModelForCausalLM ? <<< If so it would be prudent to be able to intialize a model with the pretrained components ... ie a lnaguge model and a whisper and a clip model and a xclip ? or to the effect of ... enableing for the genration of modelif these (bodys) ... ie collection of tensors whcih can be read by the imagebind library transfromer... instead of a simple tokenizer .... it would then be able to utilize the tokenizer from the llm or combine the tokenizers into a single model by adding the speicla tokens to the languge model base ? in truth this would enable the generation of even larger models and give the pretrained model a chance to find thier homes in the new body of a (imagebind) as it is the wrapper which makes the model ...(hence perhaps also including a feature to execute code on the fly and return the executing back to the model directly if enabled or ok byu the user...as this is the method as the model could generate its on code to execute in opython and return the result to the transformer to be utilized in the response it is generating .... so if the model is given functions as an input (it would also be able to run them in the back ground pre retuning the response) ... also given code by the user in chat it should be able to run the code in ipython behind the scenes and return the response (not on the system directoy(hence havign hiddon ipython repl) ..(it would still have to make the calls) ### Motivation rags internally ! usablity ! of the model wrapper to understand that each component could be trained seperate or even within the model itself ! as well as fitti…