tensor in validation_step cannot flow a gradient

[Nokimann]

I'm trying to calculate the gradient of pos from the model output during the validation.
I didn't execute trainer.fit(lit_model) Instead, I executed just trainer.validate(lit_model), overriding with

    def validation_step(self, batch, batch_idx):
        torch.set_grad_enabled(True)
        pos = batch.pos
        pos.requires_grad_(True)
        print(pos.requires_grad, 0)
        print(pos[0].requires_grad, 1)
        print(pos[:].requires_grad, 2)
        ...

I expected the results as

True 0
True 1
True 2

But I got the results,

True 0
False 1
False 2

What causes the False requires_grad?

[Nokimann]

I solved the problem by setting inference_mode=False