QCRI/Fanar-2-Oryx-IVU
LoRA-style vs Qwen/Qwen2.5-VL-7B-Instruct. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, proj, qkv, up_proj changed, in 32 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/2, Norms 0/122, MLP 96/280, Attention 64/324, lm_head 0/1
3 derivatives checked
- mradermacher/Fanar-2-Oryx-IVU-i1-GGUF: Quantization (fewer layers)
- mradermacher/Fanar-2-Oryx-IVU-GGUF: Same weights, converted (fewer layers)
- mabdulaziz499/Warraq-Arabic-HTR-7B: Adapter