Re-introduce the type-erasing tensor class.

[shoumikhin]

Summary: It's useful for cases when we want to specify the data type at runtime, e.g. create a random tensor based on some data type known at runtime only.

Reviewed By: bsoyluoglu

Differential Revision: D76908591

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/11790

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

⏳ 4 Pending, 3 Unrelated Failures

As of commit 52e5c87 with merge base 7565342 ():

FLAKY - The following jobs failed but were likely due to flakiness present on trunk:

• pull / unittest-buck / linux / linux-job (gh) (detected as infra flaky with no log or failing log classifier)
• trunk / test-arm-cortex-m-size-test / linux-job (gh) (detected as infra flaky with no log or failing log classifier)

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• pull / test-moshi-linux / linux-job (gh) (trunk failure)
  test_exported_decoder_xnnpack

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D76908591