[Fix] update rate type (#1166)

* update rate type

* update rate type

Author: Yunnglin

evalscope/perf/arguments.py

@@ -60,7 +60,7 @@ class Arguments(BaseArgument):
     parallel: Union[int, List[int]] = 1
     """Number of parallel requests."""
 
-    rate: int = -1
+    rate: float = -1
     """Rate limit for requests (default: -1, no limit)."""
 
     sleep_interval: int = 5
@@ -304,7 +304,7 @@ def add_argument(parser: argparse.ArgumentParser):
     # Performance and parallelism
     parser.add_argument('-n', '--number', type=int, default=1000, nargs='+', help='How many requests to be made')
     parser.add_argument('--parallel', type=int, default=1, nargs='+', help='Set number of concurrency requests, default 1')  # noqa: E501
-    parser.add_argument('--rate', type=int, default=-1, help='Number of requests per second. default None')
+    parser.add_argument('--rate', type=float, default=-1, help='Number of requests per second. default -1 means no rate limit')  # noqa: E501
     parser.add_argument(
         '--sleep-interval', type=int, default=5, help='Sleep interval between performance runs, in seconds. Default 5')  # noqa: E501
 

evalscope/perf/utils/benchmark_util.py

@@ -83,7 +83,7 @@ def is_embedding_or_rerank_api(api_name: str) -> bool:
 @dataclass
 class BenchmarkMetrics:
     concurrency: int = 0
-    rate: int = 0
+    rate: float = 0.0
     n_succeed_queries: int = 0
     n_failed_queries: int = 0
     total_first_chunk_latency: float = 0.0

evalscope/perf/utils/rich_display.py

@@ -68,7 +68,7 @@ def _analyze_results_embedding(all_results):
 
             summary.append([
                 str(int(concurrency)),
-                str(int(rate)) if rate != -1 else 'INF',
+                f'{rate:.2f}' if rate != -1 else 'INF',
                 f'{rps:.2f}' if rps is not None else 'N/A',
                 f'{avg_latency:.3f}' if avg_latency is not None else 'N/A',
                 f'{p99_latency:.3f}' if p99_latency is not None else 'N/A',
@@ -87,7 +87,7 @@ def _analyze_results_embedding(all_results):
             continue
 
     if summary:
-        summary.sort(key=lambda x: (int(x[0]), int(x[1]) if x[1] != 'INF' else float('inf')))
+        summary.sort(key=lambda x: (float(x[0]), float(x[1]) if x[1] != 'INF' else float('inf')))
 
     return summary, total_tokens, total_time
 
@@ -129,7 +129,7 @@ def _analyze_results_llm(all_results):
 
             summary.append([
                 str(int(concurrency)),
-                str(int(rate)) if rate != -1 else 'INF',
+                f'{rate:.2f}' if rate != -1 else 'INF',
                 f'{rps:.2f}' if rps is not None else 'N/A',
                 f'{avg_latency:.3f}' if avg_latency is not None else 'N/A',
                 f'{p99_latency:.3f}' if p99_latency is not None else 'N/A',
@@ -150,7 +150,7 @@ def _analyze_results_llm(all_results):
             continue
 
     if summary:
-        summary.sort(key=lambda x: (int(x[0]), int(x[1]) if x[1] != 'INF' else float('inf')))
+        summary.sort(key=lambda x: (float(x[0]), float(x[1]) if x[1] != 'INF' else float('inf')))
 
     return summary, total_tokens, total_time