sglangv0.5.2 & support Qwen3-Next-80B-A3B-Instruct

test/srt/parse_results.py

@@ -0,0 +1,57 @@
+import argparse
+import json
+import os
+
+import pandas as pd
+from tabulate import tabulate
+
+# Parse command-line arguments
+parser = argparse.ArgumentParser(description="Parse JSONL benchmark and summarize.")
+parser.add_argument("input_file", type=str, help="Path to input JSONL file")
+parser.add_argument(
+    "--md",
+    action="store_true",
+    help="If set, print the summary table in Markdown format (GitHub style)",
+)
+args = parser.parse_args()
+
+input_file = args.input_file
+base_name = os.path.splitext(os.path.basename(input_file))[0]
+output_file = f"{base_name}_summary.csv"
+
+fields = [
+    "max_concurrency",
+    "input_throughput",
+    "output_throughput",
+    "mean_ttft_ms",
+    "median_ttft_ms",
+    "p99_ttft_ms",
+    "mean_tpot_ms",
+    "median_tpot_ms",
+    "p99_tpot_ms",
+]
+
+# Read JSONL and parse
+results = []
+with open(input_file, "r") as f:
+    for line in f:
+        data = json.loads(line)
+        row = {field: data.get(field, None) for field in fields}
+        max_conc = data.get("max_concurrency")
+        out_tp = data.get("output_throughput")
+        row["per_user_throughput"] = out_tp / max_conc if max_conc else None
+        results.append(row)
+
+# Convert to DataFrame
+df = pd.DataFrame(results)
+
+# Save to CSV
+df.to_csv(output_file, index=False)
+print(f"\nSaved summary to: {output_file}\n")
+
+if args.md:
+    # Print Markdown table
+    print(tabulate(df, headers="keys", tablefmt="github", floatfmt=".3f"))
+else:
+    # Print ASCII table
+    print(tabulate(df, headers="keys", tablefmt="grid", floatfmt=".3f"))