geolazagr/greko
Full fine-tune (fewer layers) vs Qwen/Qwen3-VL-2B-Instruct. geolazagr/greko matches part of Qwen/Qwen3-VL-2B-Instruct (310 of its 625 tensors; the rest of Qwen/Qwen3-VL-2B-Instruct has no counterpart). On the shared part: geolazagr/greko is a GGUF (Q4_K_M) of different weights. Fine-tune: 100% of the weights changed, across 28 of 28 layers.
Tensors changed: Embeddings 1/1, Norms 111/113, MLP 84/84, Attention 112/112