Comment on SubQ: a sub-quadratic LLM with 12M-token contextparentComments−alexsubq4moWhat do you want in a whitepaper that was not in our blog post? There is time to add more before the whitepaper is released.−jazzypants4moI'm not GP, but I would want a benchmark that actually tests the entire context window. A benchmark that only tests the first 128K tokens effectively tells us nothing about how well it works at its full capacity.−alexsubq4moThat makes sense! We are working on that.
Comments
What do you want in a whitepaper that was not in our blog post? There is time to add more before the whitepaper is released.
I'm not GP, but I would want a benchmark that actually tests the entire context window. A benchmark that only tests the first 128K tokens effectively tells us nothing about how well it works at its full capacity.
That makes sense! We are working on that.