Phi-4-mini-instruct model throws access violation

[skolby42]

When I attempt using the Unsloth Phi-4-mini-instruct model from HuggingFace it throws an access violation on llama_decode in SafeLLamaContextHandle. I am using the InstructExecutor example from the documentation. Inference works with other GGUF models, so I'm wondering why this particular one doesn't? It's only 25 days old, so I don't think the model age is the issue.

public DecodeResult Decode(LLamaBatch batch)
{
	if (batch.TokenCount == 0)
	{
		return DecodeResult.Ok;
	}
	lock (GlobalInferenceLock)
	{
		LLamaNativeBatch batch2;
		using (batch.ToNativeBatch(out batch2))
		{
			return (DecodeResult)llama_decode(this, batch2); // Access violation here
		}
	}
}