Please excuse my lack of knowledge here, but if we know all of the vectors for the tokens in the cl100k_base model, why can’t we then create the embedding matrix? Is the embedding matrix not simply all of these rows?
segfault (Caleb Ditchfield )
Karma: 9
So, I’ve been a user for awhile now, but entirely lurk and don’t comment. Only recently actually made an account to vote on things .
...therefore, I have no comment karma. Thoughts on this case? I regularly vote on things normally (now that I have an account), but now that I don’t have any karma, I suppose this is my first comment, partially in purpose to get a single karma, but also to raise the case that there are probably many others like me. Maybe make this contingent on account age as well?
Could you define what you mean here by counterfactual impact?
My knowledge of the word counterfactual comes mainly from the blockchain world, where we use it in the form of “a person could do x at any time, and we wouldn’t be able to stop them, therefore x is counterfactually already true or has counterfactually already occured”