QuixiAI/MegaDolphin-120b
Partial fine-tune (fewer layers) vs dphn/dolphin-2.2-70b. dphn/dolphin-2.2-70b matches part of QuixiAI/MegaDolphin-120b (723 of its 1263 tensors; the rest of QuixiAI/MegaDolphin-120b has no counterpart). On the shared part: Partial change: 74% of the weights changed (attention, mlp, norms), in 60 of 80 layers.
Tensors changed: Embeddings 0/1, Norms 120/161, MLP 180/240, Attention 240/320, lm_head 0/1
1 derivatives checked
- mradermacher/MegaDolphin-120b-i1-GGUF: Quantization