Modular•10mo ago

Determining Tensor datatype at runtime

Hi folks! I'm working on a project where I need to extract a bunch of numeric data from a file, and the datatype for the file data is contained within the metadata for the file. I want the data within to always be read into a Tensor, but I'm not sure how to handle the datatyping at runtime, since you have to define the Tensor with a datatype at compile time. I have a feeling the answer lies in pointers, but I'm used to Python so I'm still learning about them 🙂 Grateful for any advice!

2 Replies

Michael K•10mo ago

fn do_stuff[T: DType]():
    # you have your DType as a parameter at compile time to do whatever you like
    var data = Tensor[T](1, 2)
    
fn main():
    # get your actual type name as a string from a file
    var type = 'f32'
    if type == 'f32':
        do_stuff[DType.float32]()
    else:
        do_stuff[DType.float64]()

fn do_stuff[T: DType]():
    # you have your DType as a parameter at compile time to do whatever you like
    var data = Tensor[T](1, 2)
    
fn main():
    # get your actual type name as a string from a file
    var type = 'f32'
    if type == 'f32':
        do_stuff[DType.float32]()
    else:
        do_stuff[DType.float64]()

Jake BrooksOP•10mo ago

ah of course, thank you!

Gaming

Programming

Determining Tensor datatype at runtime