The ridle json + sydney actually makes the model far more lucid than normally. I applied it on Mythomax and the answers are really good.
FPham
joined 1 year ago
I find it very strange attaching the gguf file to an exe - it's a very bad security idea (your antivirus needs to hash 10 GB file) and then on windows you still need to split it to exe and data, because the exe limit is 4GB so basically instead of llama.cpp you are now using llamafile that is llamacpp. Or am I missing something?
Geez, Alexa is in middle ages... I want fun Alexa!
Hahaha, this is totally Llama 2 with Alexa's voice.
Call it a voodoo, not an art.
And also made LLM for the users to abuse... Wait when twitter replies start sounding like Hitchhiker's guide to the galaxy and citing Vogon poetry, coz that's the "style" of Grok.
I use twitter while drinking morning coffee - it makes it 2x stronger.
I'm thinking that's probably it.
Commu - what? Blasphemy....
That looks very clean for sure.
My private finetunes are about text rewriting - input text paragraph - rewrite it in a certain style.
No 7b finetuned model can grasp the idea of submitted text in entirety, tried maybe 100 different runs. It would make a common mistake of "someone" who just scan the text quickly while also watching youtube on a phone, failing to comprehend who is who or what the paragraph is about.
13b with the same finetuning does much better - it would comprehend the relations. For example if two people are speaking, it can keep track who is who, even without mentioning it in the text.
33b - gets even further - sometimes surprise with the way it understand the text. And so the rewritten text is a mirror image of the input, just with different style
7b are impressive if you want a small local LLM to give you answers on questions, but that's probably the limit. If you want an assistant that can also do other things, then it falls short, because your instructions are not necessary understood fully.
view more: next ›
I could write a book about this.
First no model will be able to write well in style x and then in style y. Not at the size we can work with. Not at the size of ChatGPT either, although it would generally work better (it usually well approximates what the style is and falls into a Hollywood imagination of that style)
With finetuning you can make the model do one of those things.
## Instruction:
Rewrite the following text as a poem: Lister sat near the fireplace, looking at his arms which were covered in ointment. I find it hard to believe that pain can form words. Is this something internal or external? On his left arm, the message read "Dying" while on his right arm, it said "Living". He did not know what to think. A man who had been through a lot.
## Response:
Lister sat by the hearth and gazed
Upon the ointment-soaked clay
That coated his sinewy bare arms.
How could such agony frame
Such mystic characters? On one arm
The word "Dying" was graven clear;
On the other, "Living" appeared.
What could it mean? Ah! there,
A man whose life had been one tale
Of woes, of tortures, of crosses, of shame,
Who knew not what to hail as his own name.