The json is a big object and in the explanation ollama model doesn't read it correctly and gives wrong values. Meta-Llama-8B does it perfectly. Could it be the quantization?