I like this model much- it's more beautiful than most others ;-)
@DanFrederiksen Жыл бұрын
I'm guessing it's not actually slow, that's just the execution speed of the hardware you rented. I'm assuming the speed is proportional to the params in all cases. So it would be significantly faster than GPT3-175. I'm assuming it's not outperforming GPT4. I'm also assuming its knowledge is significantly smaller than GPT3 and 4. So I'm not sure it's a significantly different thing. They just ran a smaller model for longer, maybe removed some of the harder data in the training set so it's a cleaner function. Has it been limited to english or does it try to speak all languages as well?
@user-wr4yl7tx3w Жыл бұрын
Why can’t we fine tune the Falcon instruct model?
@ajaychinni3148 Жыл бұрын
We can fine tune the Falcon instruct model. I have fine tuned both of them on my custom datasets i observed the base model loss was decreasing much better than the fine tune version. I guess fine tune model has specific pattern of giving instructions if we follow the same it would be great but if you want some custom instruction styles base would be a better choice.
@jacques42 Жыл бұрын
When clcking on 'Load' for the second h2o model you mentioned, it produces a lot of errors on the right-hand-side: valueerrors and traceback errors. Does one need to further configure this model before it can load?
@haidara77 Жыл бұрын
I wonder if it would be posible to run these language models offline, so it's possible to use it in school 😅
@sabre_code Жыл бұрын
128 gb Mac can probably run it. At least theoretically
@MuhammadZahidIqbal-y2z Жыл бұрын
Does this model supports conversation history? Anyone tried?
@user-wr4yl7tx3w Жыл бұрын
Is Runpod free?
@ilianos Жыл бұрын
No, you can see the price (per hour) in the video. You need to fill up your balance (minimum $10) before you can deploy your first pod.