Alyon-AI/UMA-VLM-Engine-v1
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 k_proj, o_proj, q_proj, v_proj changed, in 28 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/2, Norms 0/122, MLP 0/280, Attention 112/324, lm_head 0/1
2 derivatives checked
- mradermacher/AQUILA-Engine-v1-i1-GGUF: Quantization (fewer layers)
- mradermacher/AQUILA-Engine-v1-GGUF: Same weights, converted (fewer layers)