Muhammadidrees/Medgamma27B
Full fine-tune (fewer layers) vs google/gemma-3-27b-pt. Muhammadidrees/Medgamma27B matches part of google/gemma-3-27b-pt (807 of its 1247 tensors; the rest of google/gemma-3-27b-pt has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 62 of 62 layers.
Tensors changed: Norms 373/373, MLP 186/186, Attention 248/248
2 derivatives checked
- mradermacher/Medgamma27B-i1-GGUF: Quantization
- mradermacher/Medgamma27B-GGUF: Quantization
2 other repos hold byte-for-byte identical weight files.