focus on rms_norm

Author: yf225

benchmarks/run.py

@@ -42,9 +42,6 @@
         "tritonbench.operators.rms_norm.operator",
         "examples.rms_norm",
         "rms_norm_tritonbench",
-        {
-            "num_inputs": 3
-        },  # TODO(yf225): reduction dim size = 8192 currently throws error
     ),
     "layer_norm": (
         "tritonbench.operators.layer_norm.operator",

benchmarks/run_input_shard.sh

@@ -8,14 +8,14 @@ OUTPUT_DIR="benchmarks_autotune_${TIMESTAMP}_input_shard_${SHARD}_of_${WORLD_SIZ
 
 KERNEL_NAME_LIST=(
     "rms_norm"
-    "layer_norm"
-    "softmax"
-    "cross_entropy"
-    "sum"
-    "jagged_mean"
-    "vector_add"
-    "embedding"
-    "vector_exp"
+    # "layer_norm"
+    # "softmax"
+    # "cross_entropy"
+    # "sum"
+    # "jagged_mean"
+    # "vector_add"
+    # "embedding"
+    # "vector_exp"
 )
 
 # Retry until success