EQUES/OpenRS3-GRPO-ja
Full fine-tune vs SakanaAI/TinySwallow-1.5B-Instruct. Fine-tune: 100% of the weights changed, across 28 of 28 layers.
Tensors changed: Embeddings 1/1, Norms 40/57, MLP 84/84, Attention 196/196
1 derivatives checked
- mradermacher/OpenRS3-GRPO-ja-GGUF: Full fine-tune