1. Update MOE; 2. Update sglang mHC; 3. Update test scripts; 4 Add new

   ops.

op_tests/test_rmsnorm2d.py → op_tests/ci_tests/test_rmsnorm2d.py

@@ -6,7 +6,7 @@ import aiter
 from aiter.test_common import checkAllclose, perftest
 from aiter import dtypes
 import argparse
-
+import pandas as pd
 
 ERROR_SUMMARY_ROWS_OUT = []
 ERROR_SUMMARY_ROWS_RES = []
@@ -97,7 +97,7 @@ def print_error_metrics(ref, pred, name, dim, dtype, ck_us=None):
 
 def print_error_summary_table(rows):
     if not rows:
-        return
+        return []
 
     cols = [
         ("case", "case"),
@@ -147,6 +147,7 @@ def print_error_summary_table(rows):
     for fr in formatted_rows:
         print(" | ".join(f"{fr[k]:<{widths[k]}}" for k, _ in cols))
 
+    return formatted_rows
 
 @perftest()
 def run_torch(input, weight, eps, residual=None):
@@ -308,5 +309,8 @@ for dtype in l_dtype:
         for n in l_n:
             test_rmsnorm2d_fuseAdd(dtype, m, n)
 
-print_error_summary_table(ERROR_SUMMARY_ROWS_OUT)
-print_error_summary_table(ERROR_SUMMARY_ROWS_RES)
+csv_rows = []
+csv_rows.extend(print_error_summary_table(ERROR_SUMMARY_ROWS_OUT))
+csv_rows.extend(print_error_summary_table(ERROR_SUMMARY_ROWS_RES))
+if csv_rows:
+    pd.DataFrame(csv_rows).to_csv("rmsnorm2d.csv", index=False)

op_tests/test_topk_plain.py → op_tests/ci_tests/test_topk_plain.py

@@ -190,5 +190,6 @@ df = pd.DataFrame(df)
 df["speedup (aiter vs torch)"] = df["time_us (torch)"] / df["time_us (aiter)"]
 df["speedup (aiter vs triton)"] = df["time_us (triton)"] / df["time_us (aiter)"]
 
+df.to_csv("topk_plain.csv",index=False)
 df_md = df.to_markdown(index=False)
 aiter.logger.info("topk_plain summary (markdown):\n%s", df_md)