Qwen/Qwen-Image-Bench

Full fine-tune vs Qwen/Qwen3.6-27B. Fine-tune: 98% of the weights changed, across 64 of 64 layers.

Tensors changed: Embeddings 1/4, MLP 192/304, Attention 352/556, lm_head 1/1, Norms 109/319

2 derivatives checked

• DawnW0lf/Qwen-Image-Bench-GGUF: Same weights, converted
• DevQuasar/Qwen.Qwen-Image-Bench-GGUF: Quantization

On the Hub · Downloads on Model Pulse