inference-optimization/Qwen3-8B-speculators.peagle-qwen3arch-ckpt4
Full fine-tune vs Qwen/Qwen3-8B. Fine-tune: 100% of the weights changed, across 4 of 4 layers.
Tensors changed: Embeddings 1/1, Norms 17/17, MLP 12/12, Attention 13/13
Full fine-tune vs Qwen/Qwen3-8B. Fine-tune: 100% of the weights changed, across 4 of 4 layers.
Tensors changed: Embeddings 1/1, Norms 17/17, MLP 12/12, Attention 13/13