NewstaR/Newstar-Qwen3-0.6B
Full fine-tune vs Qwen/Qwen3-0.6B-Base. Fine-tune: 100% of the weights changed, across 28 of 28 layers.
Tensors changed: Embeddings 1/1, Norms 86/113, MLP 84/84, Attention 112/112
2 derivatives checked
- mradermacher/Newstar-Qwen3-0.6B-GGUF: Same weights, converted
- NewstaR/Newstar-Qwen3-0.6B-KTO: LoRA-style