I found a company called goose and they had some of old BLOOM/GPT3 competitors in their arsenal of models (models such as GPT-Neo and GPT-J). But I'm looking for a company who uses BLOOM/BLOOMZ as their base model. Is there any?
I emailed goose a couple weeks ago asking if they'd add bloom. They haven't replied lol.
Not a company but I just moved https://nlp.henzi.org to BLOOMZ from BLOOM. Though disappointed in the results of continual text generation because the stop token is so easily found. In my case/usage, I am wanting to prefer open ended text generation without ending. However the "reasoning" skills of BLOOMZ is better. In that, when telling a story I noticed it was better at tracking "shopkeepers" 1, 2 and 3 in a conversation about food prices with them having varied jobs/outlooks. BLOOM generated text very well but would lose the focus / attention to the finer details at times or more easily stray. I initially was impressed with BLOOMZ for this but have suppressed selecting of the stop token when detected. This is because it will be persisted in the database and we'll start a new topic. My next step would be to remove the stop token altogether and not persist it but I'm trying to keep my backend code stable for now since it is at least working.
Here's an example story:
It wasn't so clear or settled, Einstein had his position and Newton had his. "Simple, all matter is energy - or can be" said Albert. "Yes" said Isaac Newton, "But you must remember that gravity
also produces forces, no matter the nature of the particle that causes the force". Einstein made many attempts to reconcile the two postulates, but Newton never came up with any direct solution. Einstein said: "I don't believe in the probability of things happening - although I suppose you can find some that will - which is why I didn't make the argument. But this is the problem I want to be solved and how I ought to solve it. By degrees we tend to give more thought to the part of the activity that is not related to the physical world, and to the part of the activity that is connected to it, the part that should be known to us." It is not a new idea or theory, but a new approach that, although very original, could be understood and used by all who have been influenced by the ideas of the past.
I think the question is, what are the conditions for a theory to succeed? What can it be done about it? I think that'resolving' generally means "truthfully understanding" in everything under the ground in terms of mathematical definitions.
Everything in the universe is open and causal and governed by its laws. It is the universe that produces all its phenomena, which is the most beautiful, and in this way we must accept this.
We are in this universe, as well as many other universes. But in all the other parts of that universe there is also another and higher stage. To look at it without it being known is quite different. You should have something of the original level
- of consciousness. In other words, you should be able to understand
- of consciousness in mind first. In fact, I believe that if you have to know
You can surely DM, sorry I didn't see the question!
I have to admit I've dropped to using the 560m version now because I'm out of memory at 32gb, i7. It's just an old optiplex I got a good deal on.