[CI][benchmarks] Update reporting schema for third party benchmarks (#4825)

First part of
#4824

Author: Egor-Krivov

benchmarks/third_party/e2e-flex_attention/transform_results.py

@@ -62,12 +62,13 @@ def parse_llm_log(log_file_path, tag, model, max_new_tokens, batch_size):
     # Create one row for each metric
     for metric_name, metric_value in metrics.items():
         row = {
-            'benchmark': 'e2e-flex-attention',
             'run_uuid': run_uuid,
-            'datetime': current_datetime,
+            'ts': current_datetime,
+            'benchmark_group': 'e2e-flex-attention',
+            'benchmark': 'e2e-flex-attention',
             'compiler': 'triton',
-            'metric_name': metric_name,
-            'metric_value': metric_value,
+            'value_name': metric_name + '_s',
+            'value': metric_value,
             'params': params_json,
             'tag': tag,
         }

benchmarks/third_party/liger_kernels/transform.py

@@ -18,17 +18,21 @@ def parse_args():
 def transform_df(df, tag):
     df_results = pd.DataFrame()
 
-    df = df[df["metric_name"].eq("speed") & ~df["gpu_name"].str.contains("NVIDIA")]
+    df = df[~df["gpu_name"].str.contains("NVIDIA")]
 
     if len(df) == 0:
         raise ValueError("No new results found, did all benchmarks just fail?")
 
-    df_results["benchmark"] = df["kernel_name"] + "-" + df["kernel_operation_mode"]
+    # df_results["benchmark"] = df["kernel_name"] + "-" + df["kernel_operation_mode"]
+    df_results["benchmark"] = df["kernel_name"]
+    mapping = {"speed": "_ms", "memory": "_memory_mb"}
+    df_results["value_name"] = df["kernel_operation_mode"] + df["metric_name"].map(mapping)
+    df_results["value"] = df["y_value_50"]
+    df_results["benchmark_group"] = "liger"
     df_results["run_uuid"] = uuid.uuid4().hex  # Generate a unique run ID
-    df_results["datetime"] = datetime.datetime.now()
+    df_results["ts"] = datetime.datetime.now()
     df_results["compiler"] = df["kernel_provider"]
     # Use the 50th percentile value.
-    df_results["time_ms"] = df["y_value_50"]
     df_results["comment"] = ""  # Empty comment
 
     # Create the parameters JSON, handling different x_value types correctly.