Tootfinder

"This single objective, predicting the next token, is the core training signal for a base LLM. The base model isn’t trained on factual accuracy, conversational ability, reasoning, or coding directly. It’s trained to predict the next token in massive amounts of text. Later post-training can then tune the model for instruction following, preference, safety, and conversational behavior.
... It’s called speculative decoding. A small fast model proposes several tokens ahead. The big mo…

So to follow up on this, I've caught it in action. Models, when quantized a bit, just do a bit more poorly with short contexts. Even going from f32 (as trained) to bf16 (as usually run) to q8 tends to do okay for "normal" context windows. And q4 you start feeling like "this model is a little stupid and gets stuck sometimes” (it is! It's just that it's still mostly careening about in the space of "plausible" most of the time. Not good guesswork, but still in the zone). With long contexts, the probability of parameters collapsing to zero are higher, so the more context the more likelihood you are to see brokenness.
And then at Q2 (2 bits per parameter) or Q1, the model falls apart completely. Parameters collapse to zero easily. You start seeing "all work and no play makes jack a dull boy” sorts of behavior, with intense and unscrutinized repetition, followed by a hard stop when it just stops working.
And quantization is a parameter that a model vendor can turn relatively easily. (they have to regenerate the model from the base with more quantization, but it's a data transformation on the order of running a terabyte through a straightforward and fast process, not like training).
If you have 1000 customers and enough equipment to handle the requests of 700, going from bf16 to q8 is a no-brainer. Suddenly you can handle the load and have a little spare capacity. They get worse results, probably pay the same per token (or they're on a subscription that hides the cost anyway so you are even freer to make trade-offs. There's a reason that subscription products are kinda poorly described.)
It's also possible for them to vary this across a day: use models during quieter periods? Maybe you get an instance running a bf16 quantization. If you use it during a high use period? You get a Q4 model.
Or intelligent routing is possible. No idea if anyone is doing this, but if they monitor what you send a bit, and you generally shoot for an expensive model for simple requests? They could totally substitute a highly quantized version of the model to answer the question.
There are •so many tricks• that can be pulled here. Some of them very reasonable to make, some of them treading into outright misleading or fraudulent, and it's weirdly hard to draw the line between them.

It was good, last year, when we booed the heck out of the US National Anthem. I think everyone understood what exactly people were booing. It wasn't the citizens or our many friends and family in the USA, it was the person driving us apart.
So this beautiful singing of the Canadian anthem by Buffalo fans this week before a game that didn't even feature a Canadian team is special, because it's Americans saying they still love and respect us, their friends.
The feeling is mutual.
Miss you, Americans.
#USA #Canada #NHL #CanPoli #USAPoli #Hockey #HNOM