Gryphe/Pantheon-Reasoning-31B-1.1
LoRA-style vs Gryphe/Gemma-4-31B-StyleTune. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj changed, in 60 of 60 layers; norms, biases and embeddings are untouched.
Tensors changed: lm_head 0/1, Embeddings 0/4, Norms 0/523, other 0/62, MLP 180/261, Attention 230/338
4 derivatives checked
- mradermacher/Pantheon-Reasoning-31B-1.1-i1-GGUF: Quantization (fewer layers)
- mradermacher/Pantheon-Reasoning-31B-1.1-GGUF: Quantization (fewer layers)
- Nimbz/Froopert-31B: Full fine-tune
- MRockatansky/gemma-4-31B-Mergemaxxed: LoRA-style