Minor Optimization: Replace np.empty_like(a) with np.empty(a.shape, dtype=a.dtype) after known initialization

[SaFE-APIOpt]

cascaded-generation/genbmm/opt/tvm_opt.py

Line 165 in 37c624e

c = np.empty_like(a)

Hi, thank you for the great work!

In the following code:

a = np.random.normal(size=shape).astype(np.float32)
b = np.random.normal(size=shape).astype(np.float32)
c = np.empty_like(a)

the third line creates an uninitialized array c with the same shape and dtype as a. While np.empty_like(a) is correct and readable, it performs internal introspection (via np.asarray) and metadata copying that are unnecessary here — both shape and dtype are already known and explicitly controlled in the previous line.

A more efficient alternative is:
c = np.empty(a.shape, dtype=a.dtype)
This avoids redundant checks and slightly improves performance, especially in high-throughput or repeatedly called initialization routines.