empero-ai/Qwythos-27B-v1
Full fine-tune vs Qwen/Qwen3.5-27B. Fine-tune: 92% of the weights changed, across 64 of 64 layers.
Tensors changed: lm_head 1/1, Embeddings 0/5, MLP 192/307, Attention 383/560, other 0/1, Norms 147/325
7 derivatives checked
- empero-ai/Qwythos-27B-v1-GGUF: Same weights, converted
- mradermacher/Qwythos-27B-v1-i1-GGUF: Quantization
- mradermacher/Qwythos-27B-v1-GGUF: Quantization
- Fastiraz/Qwythos-27B-v1-GGUF: Same weights, converted
- mlx-community/Qwythos-27B-v1-OptiQ-4bit: Quantization
- CocaKova/Qwythos-27B-v1-NVFP4: Quantization
- mb4063/Qwythos-27B-v1-ROCmFP4: Full fine-tune
1 other repos hold byte-for-byte identical weight files.