Zaixi/STELLA-VLM-32b
LoRA-style vs Qwen/Qwen2.5-VL-32B-Instruct. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj changed, in 64 of 64 layers; norms, biases and embeddings are untouched.
Tensors changed: Attention 256/576, MLP 192/388, Norms 0/194, Embeddings 0/2, lm_head 0/1
2 derivatives checked
- mradermacher/STELLA-VLM-32b-i1-GGUF: Quantization (fewer layers)
- mradermacher/STELLA-VLM-32b-GGUF: Quantization (fewer layers)