ehristoforu/phi-4-25b
Partial fine-tune (fewer layers) vs microsoft/phi-4. microsoft/phi-4 matches part of ehristoforu/phi-4-25b (243 of its 423 tensors; the rest of ehristoforu/phi-4-25b has no counterpart). On the shared part: Partial change: 70% of the weights changed (attention, mlp, norms), in 30 of 40 layers.
Tensors changed: Embeddings 0/1, Norms 60/81, MLP 60/80, Attention 60/80, lm_head 0/1
3 derivatives checked
- mradermacher/phi-4-25b-i1-GGUF: Quantization
- mradermacher/phi-4-25b-GGUF: Quantization
- ehristoforu/phi-4-45b: Partial fine-tune (fewer layers)