sionic-ai/comsat-embed-ja-0.3b-preview
LoRA-style vs cl-nagoya/ruri-v3-310m. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of Wi, Wo, Wqkv changed, in 25 of 25 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/2, Norms 0/50, Attention 50/50, MLP 50/50