MiniCPM4: Ultra-Efficient LLMs on End Devices

Randomly saw this -- no models yet.

[-]

EDIT:

There is an 8B model right now.

[-]

There are models now... Quite a few. Kind of overwhelming TBH!

[-]

Do we have any QAT trainer as open source code?

Their 3.3.1 Efficient Quantization-Aware Training in the technical report doesn't tell much.

[-]

How they will name their models in 2 years if those are already ultra?

[-]

no embedding models?

[-]

Looks good, I'm looking forward to it!!

[-]

openbmb my beloved.