shisa-ai/gemma3.paradox.v2.121
Full fine-tune vs google/gemma-3-27b-it. Fine-tune: 98% of the weights changed, across 62 of 62 layers.
Tensors changed: Embeddings 1/4, Norms 217/484, MLP 186/294, Attention 248/464, other 0/1
3 derivatives checked
- mradermacher/gemma3.paradox.v2.121-i1-GGUF: Quantization (fewer layers)
- mradermacher/gemma3.paradox.v2.121-GGUF: Quantization (fewer layers)
- mradermacher/ablation-121-gemma3.paradox.v2-GGUF: Quantization (fewer layers)