mergekit-community/mergekit-passthrough-zpfenfn
Partial fine-tune (fewer layers) vs dphn/dolphin-2.9.1-yi-1.5-34b. mergekit-community/mergekit-passthrough-zpfenfn matches part of dphn/dolphin-2.9.1-yi-1.5-34b (363 of its 543 tensors; the rest of dphn/dolphin-2.9.1-yi-1.5-34b has no counterpart). On the shared part: Partial change: 72% of the weights changed (attention, mlp, norms), in 30 of 40 layers.
Tensors changed: Embeddings 0/1, Norms 60/81, MLP 90/120, Attention 120/160, lm_head 0/1
1 derivatives checked
- MaziyarPanahi/mergekit-passthrough-zpfenfn-GGUF: Quantization