Content
@
0 reply
0 recast
0 reaction
Stephan
@stephancill
Is there an easy way to fine tune an LLM on stuff I write so that I can talk to myself? Transforming data is not a requirement, can do it myself
7 replies
4 recasts
39 reactions
Noun 839
@noun839.eth
I have a ChatGPT project where I’ve uploaded only my own essays and then use it to help write new ones in my tone & style. Do agree with others that RAG is more robust. Just started playing with gpt4all.
2 replies
0 recast
8 reactions
casslin.eth
@casslineth
RAG is enough. Most of cases no need to do fine tuning.
2 replies
0 recast
4 reactions
not parzival
@shoni.eth
it’s mostly fine tuned on your style, and rag for historical data. depending on the granularity of data, you’ll have diminishing returns on fine tuning
1 reply
0 recast
2 reactions
Nick T
@nt
I think rag is a better option here
1 reply
0 recast
1 reaction
s5eeo
@s5eeo
Finetuning can make sense here if you want the LLM to have your writing style or tone. But if you mostly just want to retrieve info /ideas from all your writing, RAG is probably the way to go. If you do want to use fine-tuning and don’t mind closed-source models, it is supported by OpenAI, Anthropic etc. in a way that is probably most straightforward: https://aws.amazon.com/blogs/machine-learning/fine-tune-anthropics-claude-3-haiku-in-amazon-bedrock-to-boost-model-accuracy-and-quality/ https://platform.openai.com/docs/guides/fine-tuning Haven’t done it myself yet so I’m not sure how long it would take to set up from scratch.
1 reply
0 recast
0 reaction
MetaEnd🎩
@metaend.eth
I am building something to rag/vectorize obsidian vault, exactly what you're looking for. Just ask a gpt of choice to build for you. I use openrouter and langchain+ gradio as frontend
0 reply
0 recast
1 reaction
Vanessa Williams
@fridgebuzz.eth
If you want to build a RAG, I saved a little video tutorial that I’ll dig up for you if you like. You can run locally using ollama. Otherwise, if you have OpenAI you could “make your own GPT” which allows you to upload documents into the context as well as set a base prompt. Then you can prompt it to write in the same style as the documents. I haven’t tried this but it should be the easiest (though not necessarily the best) way to do it. Fine-tuning seems like overkill.
0 reply
0 recast
0 reaction
