kyegomez / Med-PaLM

Towards Generalist Biomedical AI
https://discord.gg/qUtxnK2NMf
MIT License
283 stars 41 forks source link

Tokenizer Format Question #5

Closed rabbidave closed 11 months ago

rabbidave commented 11 months ago

The sample code uses random values to demonstrate functionality.

I'm unable to infer which tokenizer or module to use when I'm passing my own image via URL, local path, etc.

Same for defining simultaneous text input

e.g. describe any abnormalities in this image of the interior of the eye (URL to image of detaching retina)

Can you please either provide an additional example or help me understand what module I should be calling, format I should be using, etc.

Upvote & Fund

Fund with Polar

rabbidave commented 11 months ago

Reported; appreciate your diligence and specificity

kyegomez commented 11 months ago

@rabbidave Yes, the tokenizer to use should be GPTNeox, with the image embeddings passed in as special tokens.

kyegomez commented 11 months ago

@siavashk How did i plagiarize tree of thoughts? I implemented it faster than the original authors did.

And, I did copy code from Sophia because my code was worse than theirs and I didn't want people to use my bad code.

Pablo Picasso one of the best artists of all time is widely quoted as having said that “good artists borrow, great artists steal.”

Everybody steals.

I have 3,800 commits in 4 months, I regularly write thousands of lines of code a day, you personally might want to reconsider your own path of just mindlessly obeying other peoples commands and voicing an opinion with little evidence or proof

siavashk commented 11 months ago

I do not have anything else to say to you because I am not trying to convince you. I think there is enough information here for people to form their opinion.

kyegomez commented 11 months ago

@siavashk im very curious what do you think my intentions are? You know I'm doing this to advance our species right?

You're trying to discredit someone who works 24/7 to help Humanity, you should be ashamed of yourself.