collant

You should run `npx dalai serve`

It should be possible to train it on new data. You can run `finetune.py` after you prepare your dataset. You can use the Standord Alpaca method for data generation: [Data...

> But when I want to teach the model about some new scientific paper for example, I don't know what type of questions I want to ask in the future...

> Isn't there a way to train embeddings? Not sure how well it translates, but in stable diffusion, if you want it to draw an object it hasn't yet seen,...

I might have missed the point, because they use this approach to mostly direct the model in a certain style; and not to give it some new knowledge. But after...

You can also use Alpaca-lora (this repo) to generate questions and answer them: ## Question generation  ## Answer one question at a time  But I don't think the...

I can even see one small empirical evidence of new knowledge with the alpaca_data.json from Stanford. ```json { "instruction": "Discuss the causes of the Great Depression", "input": "", "output": "The...

> @collant were the screenshots from a custom gradio? looks neat. Sadly I agree that 7b (and 13b) are nowhere near sufficient for anything too serious. These are from gradio,...

> Is there a way to train from long-form text/articles/stories without using the question-answer format? > > Can I just feed in 1 full article text at a time through...

Hello, you can find this 13B one here: https://huggingface.co/samwit/alpaca13B-lora Otherwise, there is the 7B one here: https://huggingface.co/tloen/alpaca-lora-7b Please note these are LoRA models they need the base model to work....