Are you training your own Models like Llama instead of using GPT ?

Question

Maria Gonzalez · Accepted Answer

Yes, I've been trying my hand at training my own models, some of which are very comparable to Llama. Although GPT is useful, there are times when a tailored model is preferable. I can tailor the model's design and settings to my specific context and data to improve its predictive efficacy, for instance.  I get more granular control over the training process and can make any necessary adjustments to the model when I do it myself. GPT is a helpful tool, but I think you can get better, more specialized outcomes from investigating other methods of model training.

Answer

@matcha_anil @imtiyaz922 For taking notes, I highly recommend the Obsidian program. Particularly remarkable and helpful for managing extensive tasks and thoughts is its ability to link notes together and create a web of linked ideas. Also, the markdown syntax's adaptability facilitates the modification and structuring of notes. If you're looking for a powerful note-taking software that can also aid in streamlining your workflow and increasing your efficiency, I wholeheartedly endorse Obsidian. @mary_gonz8 Please correct me if im wrong because this what i did .🤔

Anil Matcha · Answer

@richard_gao2 Interesting to find what use-cases people are going to build with Llama

Richard Gao · Answer

Don't see much of a purpose for taking that effort right now

But once we have some NSFW chatbot integrations, we might decide to train our own Llama model for evoke-app.com

Hashnimo · Answer

Completely dependent on GPT, maybe until they somehow decide to API ban me.

Imtiyaz · Answer

@mary_gonz8 curious to know more about it and the use cases you are trying to train the model.

Anil Matcha · Answer

@mary_gonz8 Awesome, how do you get so much data to train your own models

Anil Matcha · Answer

@hashnimo It's scary

Alexander Miroch · Answer

Actually installed Alpaca on a linux host with 8Gb RAM. I'd say I 80% satisfied with the responses. It takes around 8 seconds to generate a decent response.

I'm using dalai, maybe it adds some processing time. Anyway the ChatGPT is not the only one now)

Are you training your own Models like Llama instead of using GPT ?

Replies