[AMD][Quantization] Add int4fp8_moe online quantization on ROCm (#7392)

Co-authored-by: Dehua Tang <dehtang@amd.com>
Co-authored-by: HAI <hixiao@gmail.com>
Co-authored-by: YC Tseng <yctseng@amd.com>

python/sglang/srt/utils/common.py

@@ -120,6 +120,12 @@ FP8_E4M3_MIN = -FP8_E4M3_MAX
 builtins.FP8_E4M3_MAX = FP8_E4M3_MAX
 builtins.FP8_E4M3_MIN = FP8_E4M3_MIN
 
+# explicitly use pure text format, with a newline at the end
+# this makes it impossible to see the animation in the progress bar
+# but will avoid messing up with ray or multiprocessing, which wraps
+# each line of output with some prefix.
+BAR_FORMAT = "{desc}: {percentage:3.0f}% Completed | {n_fmt}/{total_fmt} [{elapsed}<{remaining}, {rate_fmt}]\n"  # noqa: E501
+
 
 @lru_cache(maxsize=1)
 def is_cuda():