v000000/Psyonic-Rose-20B-Higher-Quality

Full fine-tune vs tavtav/Rose-20B. Fine-tune: 100% of the weights changed, across 62 of 62 layers.

Tensors changed: Embeddings 1/1, Norms 125/125, MLP 186/186, Attention 248/248, lm_head 1/1

2 derivatives checked