[d416]

Are the Q* models open source with open data? Asking because other than the known tianamen limitation, prompting these models for any advice about acts against authority in a corporate setting yields strong recommendations of compliance. Haven’t explored this deeply, but it was enough of a red flag to limit use to coding/logic tasks.

[londons_explore]

Model architecture and weights are open source.

Training data is not (but I don't think anyone in this game has fully open training data)

[marci]

Here you'll find some of the most advanced, fully open (models, data, code, logs, checkpoints...): https://allenai.org/blog/olmo2

[tokioyoyo]

Nobody wants to be sued to oblivion.

[hustwindmaple1]

There are a couple, i.e., OLMo 2

[owenpalmer]

Q* was the codename for a model at OpenAI that was eventually released under the name o1. This model is something different. QwQ and QvQ are open source models made by the Qwen Team.