cerno-4b-v1.2 + Qwen3.5-4B vision tower: Zils d3-mini soup body
This checkpoint combines two sources:
- Language model. The weights of apertixai/cerno-4b-v1.2 at revision
e14dec9894c5afb761f90926f1c19b19938a271c. They are unchanged, all 426 model.language_model.* tensors (Apache-2.0).
- Vision tower. The 297
model.visual.* tensors of Qwen/Qwen3.5-4B at revision 851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a, copied unchanged (Apache-2.0).
It is published as a body for the Zils (Bittensor subnet 127) d3-mini soup pool, whose body layout is the full Qwen3.5 checkpoint (language plus vision).
Changes from the sources (Apache-2.0 §4(b)):
- The text-only cerno export was completed with the unchanged Qwen3.5-4B vision tower, written to
model-00002-of-00002.safetensors.
- The original
model.safetensors is unchanged and renamed model-00001-of-00002.safetensors.
- A new
model.safetensors.index.json maps both shards.
config.json is the Qwen3.5-4B parent's.
zils-owner.json names the Bittensor hotkey that committed this body.
All credit goes to the original authors; see the cerno model card for usage and evaluation.
Licence: Apache-2.0; LICENSE is the Apache-2.0 text shipped with Qwen3.5-4B.