Trying something new, going to pin this thread as a place for beginners to ask what may or may not be stupid questions, to encourage both the asking and answering.
Depending on activity level I'll either make a new one once in awhile or I'll just leave this one up forever to be a place to learn and ask.
When asking a question, try to make it clear what your current knowledge level is and where you may have gaps, should help people provide more useful concise answers!
Is there a way to make my LLM permanently 'internalize' the subtle changes in behavior and important information in a more permanent way that doesn't rely on context input?
From what I understand most of the change over time happens with the processing of context. The gradual increase in context as the conversation goes on combined with the core initial system prompts are the main factors in how an LLM will behave/respond. When context is wiped and system prompt is defaulted the personality and any information gained is completely erased.
Have you tried the guide on AMD's site? It looks like it's for Windows, and I don't know what you're running. Plus, I use Ollama, so I probably can't be of much help.
For programing, my favorite is Dolphin-Mixtral, but I've had good results with Dolphin-Mistral and Llama2.
Where is the sweet spot for running CPU bound models? I've just started playing with llama.cpp but the big models do make the cores work pretty hard. Should I look at using quantisation or more fine tuned models for the tasks I care about (developer assistance mainly).
If you're using llama.cpp chances are you're already using a quantized model, if not then yes you should be. Unfortunately without crazy fast ram you're basically limited to 7B models if you want any amount of speed (5-10 tokens/s)
Is there a standard for the suffixes? For example the OpenLlama models here: https://huggingface.co/SlyEcho/open_llama_7b_v2_gguf/tree/main have qN and and then a mix of K, M, 0 and 1 suffixes. The q I assume is the quantisation level but measured how? Does q2 mean t 2bits per weight? That seems very small - and what is it fixed float, integers?
Do you usually have some other front-end over the model? I can run llama.cpp directly in interactive mode but the results are a little underwhelming. However there seem to be various front ends that get better results? Is this down to better prompting and parameter control? I've seen temperature mentioned in relation to ChatGPT but I have no idea what rope and yarn factors are for?
I use text-generation-webui mostly. If you're only using GGUF files (llama.cpp), koboldcpp is a really good option
A lot of it is the automatic prompt formatting, there's probably like 5-10 specific formats that are used, and using the right one for your model is very important to achieve optimal output. TheBloke usually lists the prompt format in his model card which is handy
Rope and yarn refer to extending the default context of a model through hacky (but functional) methods and probably deserve their own write up
I have two 3090 Turbo GPUs and it seems like oobabooga doesn't split the load between the two cards when I try to run TheBloke/dolphin-2.7-mixtral-8x7b-AWQ.
Does anyone know how to make text generation webui use both cards? Do I need an nvlink between the two cards?
You shouldn't need nvlink, I'm wondering if it's something to do with AWQ since I know that exllamav2 and llama.cpp both support splitting in oobabooga
It works ok for the most part.
The problem i have with it is that inline completion is more annoying then helpful, because the AI only sees the last few lines that you wrote and therefore does not know the larger context of the project.
I also found this project, it looks promising.
Has anyone tested it?
Can you separate the server from the client?
Thanks for the suggestion, I tried it and the diff view is very good.
The setup was not really easy for my local models, but after i set it up, it was really fast.
The biggest problem with the tool is that the open source models are not that good, i tried if it could fix a bug in my code and it was only able to make it worse.
On a more positive note, you at least do not need to copy all text over to another window and it is great for generating boilerplate code nearly flawlessly every time.