Zils d3-mini soup body: BaseDecision-Dyno (LoRA merged)

This repository re-packages onlyaady/BaseDecision-Dyno at revision 3f6690611d8bdc20ac9c9c627ef8db4f254a0612 as a soup body for the Zils subnet (Bittensor SN127), pool d3-mini. The source stores its LoRA unmerged; here each adapted weight is merged the standard way, W = base + (lora_alpha / r) * B @ A with lora_alpha=32, r=16, computed in fp32 and stored in bf16. All other language tensors are the source's, unchanged. The source's own readout head is not included. The source has no vision tower; the one here is Qwen/Qwen3.5-4B's at revision 851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a, unchanged, as the Zils body layout requires.

All credit for the weights belongs to the original authors; redistributed under their apache-2.0 licence.

Downloads last month

-

Downloads are not tracked for this model. How to track
Safetensors
Model size
5B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for subs-king/zils-mini-basedecision-dyno

Finetuned
Qwen/Qwen3.5-4B
Finetuned
(1)
this model