Diffusers Example quantize.py: --calib-size help text is misleading (acts as number of samples/prompts, not “calibration steps”). Suggest clearer name/alias and ceil batching.

The CLI flag --calib-size is documented as:

“Number of calibration steps.”

But in the script it’s used as the number of calibration samples/prompts, then divided by --batch-size to compute how many batches to run once. This is easy to misinterpret as “steps/epochs.”

Concretely, the code does:
• args.calib_size = args.calib_size // args.batch_size
• do_calibrate(...) iterates over batches of prompts and breaks when i >= calib_size (so: one pass, limited to that many batches).

Result: --calib-size 128 --batch-size 2 runs 64 batches × 2 prompts = 128 images total, not “128 calibration steps.”

This confused me because “steps” usually implies repeated passes/epochs, not “samples.”

FWIW, the Model Optimizer docs describe PTQ calibration as running on a small set of samples (typically 128–512), which matches the implementation, so the help text should say “samples/prompts,” not “steps.” 

Proposed fix (backward-compatible)
Clarify the help text and add an alias that reflects the actual meaning:

$-$ parser.add_argument("--calib-size", type=int, default=128,
$-$ help="Number of calibration steps.")
$+$ parser.add_argument("--calib-size", "--calib-samples", dest="calib_size",
$+$ type=int, default=128,
$+$ help="Number of calibration samples (prompts/images). "
$+$ "Internally divided by --batch-size to compute how many batches to run once.")

Why this matters
• Prevents users from over- or under-estimating calibration runtime (especially at high resolutions/long --n-steps).
• Aligns the CLI with the docs’ guidance that PTQ uses a small number of samples (128–512), not multiple epochs. 

Thanks for considering! Happy to submit a PR with the above diff if that helps.