improve quantization

hf_to_nvfp4_safetensors.py

@@ -291,6 +291,7 @@ def parse_args(argv: list[str] | None = None) -> argparse.Namespace:
                         help="Learned-rounding iterations per matrix (convert_to_quant default: 4000)")
     parser.add_argument("--lr", type=float, default=1.0, help="Initial learning rate")
     parser.add_argument("--lr-min", type=float, default=1e-8, help="Lowest learning rate. The flat-loss check starts here (default: 1e-8).")
+    parser.add_argument("--lr-patience", type=int, default=5, help="Patience for learning rate decay (default: 5)")
     parser.add_argument("--lr-schedule", default="plateau", choices=("plateau", "exponential", "adaptive"))
     parser.add_argument(
         "--scale-optimization", default=DEFAULT_SCALE_OPTIMIZATION, choices=("fixed", "iterative", "joint"),