After Anthropic, Moonshot is another model provider who restricts tweaking of sampling parameters. I do like the idea of the vendor verifier, though.

▲

charcircuit 4 days ago | parent | next [-]

If the post training is done with specific sampling parameters it would make sense to only use the parameters it was trained with.

▲

curioussquirrel 4 days ago | parent [-]

Yes, but post training cannot possibly account for all possible use cases. Sane defaults are fine, you can't really do much about sampling parameters in chatbots and coding harnesses anyway. And when making an API call, you have to actively change the parameter in your payload. I don't believe there's any real risk.

	▲	charcircuit 3 days ago \| parent [-]
		The risk is that people tweak it, potentially on accident, and then think the model is bad instead of understand they are using it wrong. This causes potential reputational damage by exposing the control.

▲

throwa356262 4 days ago | parent | prev [-]

What does "restricts tweaking of sampling parameters" mean?

	▲	yorwba 4 days ago \| parent [-]
		"enforcing Temperature=1.0 and TopP=0.95 in Thinking mode, with mandatory validation that thinking content is correctly passed back."