Best open source llm (from ollama) for GraphRag

[MisterAndry]

I'm curious, what open source models are you using? With which models is indexing most often successful?
From my practice, llama 3.1 70b 8 bit in half of the cases indexing does not reach the finish line (sometimes the model does not follow the prompt and the output format is wrong, hence the graph is not built).
qwen2 72b 8 bit, even less often it finishes successfully.

[MisterAndry]

found the same discussion