dalle-2-preview VS sentencepiece

We believe that many decisions about our defaults and hard bounds should be made collectively, and while practical implementation is a challenge, we aim to include as many perspectives as possible. As a starting point, we’ve sought external input on our technology in the form of red teaming. We also recently began soliciting public input on AI in education (one particularly important context in which our technology is being deployed).

From this OpenAI document:

It's also of note that you can't sell a game using this method, as Dalle-2's terms of service prevent use in commercial projects. It's hard to justify rate of return considering you can only ever give it away for free, and even in that case there are some uncertain legal elements regarding copyright and the images that are used to train the dataset.

On their GitHub https://github.com/openai/dalle-2-preview/blob/main/system-card.md I can only see references to v1.

"b) Exploration of the bolded part of OpenAI's comment "Each generated image includes a signature in the lower right corner, with the goal of indicating when DALL·E 2 helped generate a certain image." (source)." (source link: https://github.com/openai/dalle-2-preview/blob/main/system-c...)

I feel the DALL-E 2 watermark signature could be a seed or something.

DALL-E's docs for example mention it can output whole copyrighted logos and characters[1] and understands it's possible to generate human faces that are bear the likeness of those in the training data. We've also seen people recently critique Stable Diffusion's output for attempting to recreate artists' signatures that came from the commercial trained data.

That said by a certain point the kinks will be ironed out and likely skirt around such issues by only incorporating/manipulating just enough to be considered fair use and creative transformation.

[1] "The model can generate known entities including trademarked logos and copyrighted characters." https://github.com/openai/dalle-2-preview/blob/main/system-c...

Some types of content were purposely excluded from the training dataset(s) (source).

you need to train the model on 1 trillion tokens (https://platform.openai.com/tokenizer https://github.com/google/sentencepiece) anyways for it to get reasoning capacities, which it feels very unlikely that your data is that much.

I'm highly skeptical that you have enough data to pretrain if you don't have enough data to fine tune.

fine tuning + vector search + prompting of as much stuff as you can, on a LLM like palm2 or gpt4 is what I would do. otherwise you can use falcon 40B ofc.

maybe I should charge for this ahah

a) Comparison with SentencePiece tokenizer with comparable settings (It can also ignore word-boundaries and create phrase tokens)

LLaMA uses the sentencepiece tokenizer: https://github.com/google/sentencepiece

Besides, are you familiar with SentencePiece? What you are doing looks very similar (generate a large vocab, prune worst token until vocab size is reached), only the token selection criterion is different. It's also purely data driven in the sense that there are no assumption specific to language (and it can optionally segment across whitespace, as you are doing).

Hi, I'm studying the implementation of encode and decode functions for Google's SentencePiece tokenizer.

I think many LMs nowadays use unicode tokenizers, that are not tied to specific languages. E.g. sentencepiece is the most popular one: https://github.com/google/sentencepiece

If you are interested in implementing LLaMA yourself or learning, I noticed that the reference code by Facebook is one of the cleaner, easier to read ML code I've seen in a while. https://github.com/facebookresearch/llama/blob/main/llama/mo... It's about 200 lines long. You probably do need a bit of knowledge to understand what you are reading but I was pleasantly surprised.

For example in comparison, StableDiffusion torch code in diffusers and transformers Python libraries has lots of conditionals, experiments etc. that are not being used that can make it hard to follow what is going on.

Last weekend I got the "main loop" of the transformer working in pure CPU Rust code, following the reference code. My crappy code is just very very slow as I focused on getting it to run, not making it fast. The tokenizer uses some Google thing https://github.com/google/sentencepiece but luckily for inference it seems that you just need to be able to parse the tokenizer model file and not understand how it was created; I was able to strip out the protobuf files from that repository and add it to Rust and read the tokens.

I am optimistic that someone makes a high quality CPU or some CPU+GPU+SSD combination thingmaling that will make it somewhat practical to run even the large LLM models without needing an A100 or two.