<feed xmlns='http://www.w3.org/2005/Atom'>
<title>boris-2, branch main</title>
<subtitle>Boris-2: training code for the 144M Boris-2 language model (OpenCerebral)
</subtitle>
<id>https://git.kunix.org/boris-2/atom?h=main</id>
<link rel='self' href='https://git.kunix.org/boris-2/atom?h=main'/>
<link rel='alternate' type='text/html' href='https://git.kunix.org/boris-2/'/>
<updated>2026-11-21T17:36:58Z</updated>
<entry>
<title>mix.md: ClimbMix was not decontaminated</title>
<updated>2026-11-21T17:36:58Z</updated>
<author>
<name>jdj</name>
<email>jdj@kunix.org</email>
</author>
<published>2026-11-21T17:36:58Z</published>
<link rel='alternate' type='text/html' href='https://git.kunix.org/boris-2/commit/?id=0ebd25864322ac8706fbaf1e0acb6124b36689d8'/>
<id>urn:sha1:0ebd25864322ac8706fbaf1e0acb6124b36689d8</id>
<content type='text'>
The decontamination section said every source passes the 13-gram filter.
ClimbMix was tokenized in bulk shards outside prepare.py and was not
filtered, as the model card already states. Add the per-source dropped
counts for the filtered sources.
</content>
</entry>
<entry>
<title>Add mix.md: the complete data mixture</title>
<updated>2026-11-18T22:52:51Z</updated>
<author>
<name>jdj</name>
<email>jdj@kunix.org</email>
</author>
<published>2026-11-18T22:52:51Z</published>
<link rel='alternate' type='text/html' href='https://git.kunix.org/boris-2/commit/?id=79aff457f6b3a9cf2dd797d1355cb7024b01117e'/>
<id>urn:sha1:79aff457f6b3a9cf2dd797d1355cb7024b01117e</id>
<content type='text'>
Every source with its dataset, subsets, filters and license; the
draw-weight schedule of every phase; batch, context and final stages.
</content>
</entry>
<entry>
<title>Boris-2 training code</title>
<updated>2026-11-18T22:47:33Z</updated>
<author>
<name>jdj</name>
<email>jdj@kunix.org</email>
</author>
<published>2026-11-18T22:47:33Z</published>
<link rel='alternate' type='text/html' href='https://git.kunix.org/boris-2/commit/?id=5d2e4252e695644c34be605b44a14bd80a98a0a6'/>
<id>urn:sha1:5d2e4252e695644c34be605b44a14bd80a98a0a6</id>
<content type='text'>
The code, configuration and data filters used to train Boris-2 (144M,
200B tokens, single V100): model, training loop, data preparation,
n-gram table, V100 kernels, final stages and the exact run configs.
</content>
</entry>
</feed>
