Kolibri – Tech Report [pdf]
aleph-alpha.com107 points by yu3zhou4 5 hours ago
107 points by yu3zhou4 5 hours ago
Since it's a PDF, here's an excerpt from the summary:
> This report introduces Kolibri, an English–German Mixture-of-Experts transformer with 78.1B total parameters and 3.46B active parameters per token, released as open weights under the Apache 2.0 licence.
> We built Kolibri for sovereign and specialised deployment, with a particular focus on German and on regulated domains such as public administration, industry, and aerospace. To support these settings, we control the full model-development process, including data, architecture, training infrastructure, post-training, and evaluation.
It’s not a dupe, it’s a full training and infra report
The criterion that decides dupeness is not whether the URLs are identical, but whether the new submission can support a substantially different discussion.
The report is certainly useful! but there is no way that the two threads would remain partitioned, with one about the model in general and another about this report in particular.
Therefore the later submission, although the link is different, needs to be treated as a dupe. The solution is to merge the threads while including the new link prominently in the other thread, so readers don't lose access to it. I've done that now by putting https://aleph-alpha.com/downloads/tech-report.pdf into the toptext of the main thread.
Btw, this is the the same logic that we applied in merging https://news.ycombinator.com/item?id=49943034 into the primary thread. Both links are useful, but the discussions were so overlapping that they needed to be merged.