mradermacher/SVILA-1-12B-GGUF
Full fine-tune (fewer layers) vs GaMS-Beta/SVILA-1-12B. mradermacher/SVILA-1-12B-GGUF matches part of GaMS-Beta/SVILA-1-12B (627 of its 1066 tensors; the rest of GaMS-Beta/SVILA-1-12B has no counterpart). On the shared part: mradermacher/SVILA-1-12B-GGUF is a GGUF (Q8_0) of different weights. Fine-tune: 100% of the weights changed, across 48 of 48 layers.
Tensors changed: Embeddings 1/1, Norms 121/289, MLP 144/144, Attention 192/192, lm_head 1/1