Every fine-tune on the Hub, layer by layer

Every fine-tune on the Hugging Face Hub as a bar of its layers, measured from the weights against the parent it declares: which layers moved, whether a quant is really a quant, which repos hold the same weights. 120,976 pairs checked; paste any model to compare it.

• Qwen/Qwen3.8-27B: 1,260 derivatives checked
• black-forest-labs/FLUX.1-dev: 852 derivatives checked
• Qwen/Qwen3.6-35B-A3B: 644 derivatives checked
• meta-llama/Llama-3.1-8B-Instruct: 634 derivatives checked
• Qwen/Qwen3.5-9B: 606 derivatives checked
• Qwen/Qwen3.5-4B: 583 derivatives checked
• Qwen/Qwen3-8B: 574 derivatives checked
• Qwen/Qwen2.5-7B-Instruct: 566 derivatives checked
• Qwen/Qwen3-4B-Instruct-2507: 486 derivatives checked
• Qwen/Qwen3.6-27B: 437 derivatives checked
• Qwen/Qwen3-4B: 421 derivatives checked
• Qwen/Qwen3-0.6B: 405 derivatives checked
• Qwen/Qwen3-1.7B: 380 derivatives checked
• stabilityai/stable-diffusion-xl-base-1.0: 370 derivatives checked
• Qwen/Qwen2.5-1.5B-Instruct: 367 derivatives checked
• Qwen/Qwen3.8-Flash-Next: 334 derivatives checked
• Qwen/Qwen2.5-VL-7B-Instruct: 333 derivatives checked
• Qwen/Qwen2.5-3B-Instruct: 326 derivatives checked
• Qwen/Qwen3.5-0.8B: 306 derivatives checked
• Qwen/Qwen2.5-0.5B-Instruct: 301 derivatives checked