Vortex5/SpicyFlyRP-22B
Full fine-tune vs anthracite-org/magnum-v4-22b. Fine-tune: 98% of the weights changed, across 56 of 56 layers.
Tensors changed: Embeddings 1/1, Norms 103/113, MLP 165/168, Attention 220/224, lm_head 1/1
1 derivatives checked
- mradermacher/SpicyFlyRP-22B-i1-GGUF: Quantization