inference-optimization/Qwen3-8B-from-Qwen3-8B_regen-speculators.eagle3-qwen3arch-ckpt1

Partial fine-tune vs Qwen/Qwen3-8B. Partial change: 21% of the weights changed (attention, mlp, norms), in 1 of 1 layers.

Tensors changed: Embeddings 0/1, Norms 5/5, MLP 3/3, Attention 1/1

On the Hub · Downloads on Model Pulse