This issue tracks progress and features related to make_tensor. This is in the direction of exposing it to the users, and hence a tracker to invite discussions/inputs as well as document progress.

Features Tracker

• Fix inputs low/high given: don't clamp to default values when given valid values. (being pursued in Fixing user inputs for low, high in make_tensor #61108)
• Investigate tensor generation strategy. (being pursued in Fixing user inputs for low, high in make_tensor #61108)
• Generic exclude_values and include_values for tensor generation.
• Add include_percent like argument to populate the output with given percent of include_values from the 3rd task.
• Add include_lhs (default True) and include_rhs (default False) to the function.
• Document make_tensor

Description

make_tensor is planned (and expected) to go public to the user API in coming releases. The idea behind make_tensor is to allow generation of tensors with random (currently: uniformly distributed) values for given dtype and device. Keeping the user's requirements in mind as well as the extensive use of make_tensor in the PyTorch test suite, this issue is supposed to document the progress as well as invite discussions on the features listed above.

cc: @mruberry @pmeier