Tensorflow raw_ops print function gives error: NotFound: No attr named 'U' in NodeDef

1

Using rust crate tensorflow version = "0.21", with features = ["eager","tensorflow_gpu"], the following code compiles cleanly but gives this weird runtime error: NotFound: No attr named 'U' in NodeDef.

// Create an eager execution context with the specified options
let options = ContextOptions::new();
let ctx = Context::new(options).unwrap();
let oth = original_tensor.to_handle(&ctx).unwrap();
let otr = orig_tensor_rejoined.to_handle(&ctx).unwrap();
unsafe {
    let raw_sum: Tensor<f32> = add(&ctx, &oth, &otr)  
        .unwrap()
        .resolve()
        .unwrap()  // turning the resulting ReadonlyTensor, here, back into a tensor is unsafe
        .into_tensor();

    let r_handle = raw_sum.into_handle(&ctx).unwrap();

    // Use the print op to print the tensor. This part is inside the unsafe block only due to scope of
    // raw_sum
    println!("Raw sum is:");
    print(&ctx, &r_handle, &oth).unwrap();
    }

The documentation for the C++ api seems to suggest that while &r_handle is a reference to a TensorHandle (as I have), the &oth should be a reference (pointer) to a list of tensors to print out when op is evaluated (see docs here: tensorflow::ops::Print Class Reference  |  TensorFlow v2.16.1) but this is not at all clear to me (what is the difference and why).

In my code, if I try to replace &oth with some kind of reference to an array of handles, the code won't compile. Is this a bug in the tensorflow crate or am I doing something wrong?

2

This topic was automatically closed 90 days after the last reply. We invite you to open a new topic if you have further questions or comments.