jialinyyzz/Qwen3.8-27B-DSpark-drafter

Full fine-tune vs RadixArk/Qwen3.8-27B-DSpark. Fine-tune: 100% of the weights changed, across 5 of 5 layers.

Tensors changed: other 5/5, Norms 22/22, Attention 10/10

2 other repos hold byte-for-byte identical weight files.

On the Hub · Downloads on Model Pulse