> Scaling post-training is all we did for GLM-5.3.
Love this opening line. And wow, great results.
> As agent capability improves, much of the difficulty in scaling post-training moves from the model to the environment.
> Scaling post-training is all we did for GLM-5.3.
Love this opening line. And wow, great results.
> As agent capability improves, much of the difficulty in scaling post-training moves from the model to the environment.
What actually is "scaling post-training"?
More RLVR. Give it verifiable problems, if it doesn't find a solution move on, if it does, use that as a reward signal.
Can’t this be extended quite far? Use a cerebras-served model, use verification techniques to generate and solve millions of problems and then use that as training?
This isn't latency bound, it is trivially parallelize. So you want to run it on the most efficient compute you have, not the fastest.
OpenAI was bragging the other day about how HLE on Cerebras only took 11 hours. But you're saying if they just spun up a separate cheapo VM for each task, it would have finished faster and cheaper?
That’s the whole point, just cost and compute limitations in your way (mostly).
does this suggest 5.3 is the same # of parameters as 5.2?
which is the bigger headline that people don't realize. this is 744b and its head to head with Kimi K3 (2.8T), smashes DS v4 pro (1.5T). even Opus and Sol are rumored to be 1.5T+ this is half the size!
in terms of performance the formula seems to be : dense parameters = sqrt(total*active)
You do need to compare active parameter too though. The total size isn’t a reliable indicator anymore
It's the same pre-training, they are just adding more (+ better) SFT, RL, etc. (post-training). Model internal knowledge cut-off is still the same.
It seems we are doing pre-training every 6 months, and post-training every 4-8 weeks now.
“Today we are releasing GLM-5.3. It uses the same base model as GLM-5.2 — every gain comes from post-training.“
I was asking if this implies that 5.3 has the same number of parameters as 5.2. I can, in fact, read. What I didn't do is understand the implication of that statement. Thank you for your copy/paste service.