QuixiAI/Qwen3-72B-Synthesis
Partial fine-tune vs Qwen/Qwen2.5-72B-Instruct. Partial change: 80% of the weights changed (attention, mlp, norms), in 64 of 80 layers.
Tensors changed: Norms 129/161, MLP 192/240, Attention 256/320
4 derivatives checked
- mradermacher/Qwen3-72B-Instruct-GGUF: Full fine-tune
- mradermacher/Qwen3-72B-Instruct-i1-GGUF: Full fine-tune
- mradermacher/Qwen3-72B-Synthesis-i1-GGUF: Quantization
- mradermacher/Qwen3-72B-Synthesis-GGUF: Quantization