beyoru/EvolLLM
Full fine-tune vs Qwen/Qwen3-4B-Instruct-2507. Fine-tune: 100% of the weights changed, across 36 of 36 layers.
Tensors changed: Embeddings 1/1, Norms 145/145, MLP 108/108, Attention 144/144
2 derivatives checked
- beyoru/MinCoder-4B-Expert: Full fine-tune
- beyoru/Luna-7B-A4B: MoE built from dense models