All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Openai equivalent or some easy diy self made gpt ?
I have been trying to do something very simple with openai's api. I am asking it to generate text/product description in multiple languages using product attributes i provide it.
Most of the times it doesnt follow instruction, sometimes the returned json has errors, sometimes text is too short despite minimum length instruction, sometimes its just bad gateway error,etc. Around 30% of my tokens go wasted but I still have to pay for them.
Since I have a very small and targetted use case can anyone suggest if there is an easy way to train a self hosted ai solution which does just that ie create descriptions using attributes i provide. I have a sample of around 10k attributes and their corresponding descriptions. If creating something like this is no easy feat then can anyone think of other options, is google cloud's api a viable alternative ?
Comments
Fine-tuning on openAI?
I have looked at it but i dont understand how it cud work?
Altman got fired. Blame on this.
So you make the chats yourself, here is an example:
{"messages": [{"role": "user", "content": "Here is a product description: bla bla bla bla"}, {"role": "assistant", "content": "THE DESCRIPTION OF THE PRODUCT"}]}
Then add new lines for each product, then feed them into fine-tuning. DM me if you need help [this would also be my first time doing this, but this seems fun enough]
Kind of. Meanwhile they seem to want to have him back.
https://github.com/lm-sys/FastChat
Supports same API commands/structure as openai.
Basically a self hosted replacement to GPT-4 which is compatible with openai formats and API and third party tools.
Install this https://github.com/oobabooga/text-generation-webui
Then load a model using cpp with like this model: https://huggingface.co/TheBloke/dolphin-2.2.1-mistral-7B-GGUF
Then ask it your question.
All run locally there u go.
For self-hosted, you might want to try something based on Meta's LLaMa, but you'll need a GPU with a decent amount of VRAM to get the best performance from it.
Large language models are not good at counting. In fact, they're not good at anything involving numbers. They're not designed for that. You need to understand the limitations of the model you're using to get the most out of it.