Plumb output changes though progress and summary

Author: sjmonson

src/guidellm/benchmark/benchmark.py

@@ -277,7 +277,7 @@ class Benchmark(StandardBaseModel):
     Specific benchmarker implementations should extend this model to include
     additional information or metadata as needed.
 
-    Note, requests_per_second and requests_concurrency are kept at this level
+    Note, requests_per_second and request_concurrency are kept at this level
     and are expected to be populated by the subclass implementation to ensure
     the logic for Profiles can include more complicated logic for determining
     what rates and concurrency values to use for subsequent strategies.

src/guidellm/benchmark/benchmarker.py

@@ -235,8 +235,8 @@ async def run(
 
             benchmark: BENCH = aggregator.compile()
             profile.completed_strategy(
-                average_rate=benchmark.requests_per_second.successful.mean,
-                average_concurrency=benchmark.requests_concurrency.successful.mean,
+                average_rate=benchmark.metrics.requests_per_second.successful.mean,
+                average_concurrency=benchmark.metrics.request_concurrency.successful.mean,
             )
 
             yield BenchmarkerResult(

src/guidellm/benchmark/output.py

@@ -258,29 +258,29 @@ def print_benchmarks_info(self):
                     f"{datetime.fromtimestamp(benchmark.end_time).strftime("%H:%M:%S")}",
                     f"{(benchmark.end_time - benchmark.start_time):.1f}",
                     (
-                        f"{benchmark.successful_total:>5} / "
-                        f"{benchmark.incomplete_total} / "
-                        f"{benchmark.errored_total}"
+                        f"{benchmark.total_count.successful:>5} / "
+                        f"{benchmark.total_count.incomplete} / "
+                        f"{benchmark.total_count.errored}"
                     ),
                     (
-                        f"{benchmark.prompt_token_count.successful.mean:>5.1f} / "
-                        f"{benchmark.prompt_token_count.incomplete.mean:.1f} / "
-                        f"{benchmark.prompt_token_count.errored.mean:.1f}"
+                        f"{benchmark.metrics.prompt_token_count.successful.mean:>5.1f} / "
+                        f"{benchmark.metrics.prompt_token_count.incomplete.mean:.1f} / "
+                        f"{benchmark.metrics.prompt_token_count.errored.mean:.1f}"
                     ),
                     (
-                        f"{benchmark.output_token_count.successful.mean:>5.1f} / "
-                        f"{benchmark.output_token_count.incomplete.mean:.1f} / "
-                        f"{benchmark.output_token_count.errored.mean:.1f}"
+                        f"{benchmark.metrics.output_token_count.successful.mean:>5.1f} / "
+                        f"{benchmark.metrics.output_token_count.incomplete.mean:.1f} / "
+                        f"{benchmark.metrics.output_token_count.errored.mean:.1f}"
                     ),
                     (
-                        f"{benchmark.prompt_token_count.successful.total_sum:>6.0f} / "
-                        f"{benchmark.prompt_token_count.incomplete.total_sum:.0f} / "
-                        f"{benchmark.prompt_token_count.errored.total_sum:.0f}"
+                        f"{benchmark.metrics.prompt_token_count.successful.total_sum:>6.0f} / "
+                        f"{benchmark.metrics.prompt_token_count.incomplete.total_sum:.0f} / "
+                        f"{benchmark.metrics.prompt_token_count.errored.total_sum:.0f}"
                     ),
                     (
-                        f"{benchmark.output_token_count.successful.total_sum:>6.0f} / "
-                        f"{benchmark.output_token_count.incomplete.total_sum:.0f} / "
-                        f"{benchmark.output_token_count.errored.total_sum:.0f}"
+                        f"{benchmark.metrics.output_token_count.successful.total_sum:>6.0f} / "
+                        f"{benchmark.metrics.output_token_count.incomplete.total_sum:.0f} / "
+                        f"{benchmark.metrics.output_token_count.errored.total_sum:.0f}"
                     ),
                 ]
             )
@@ -313,29 +313,29 @@ def print_benchmarks_stats(self):
             rows.append(
                 [
                     strategy_display_str(benchmark.args.strategy),
-                    f"{benchmark.requests_per_second.successful.mean:.2f}",
-                    f"{benchmark.requests_concurrency.successful.mean:.2f}",
-                    f"{benchmark.output_tokens_per_second.total.mean:.1f}",
-                    f"{benchmark.tokens_per_second.total.mean:.1f}",
+                    f"{benchmark.metrics.requests_per_second.successful.mean:.2f}",
+                    f"{benchmark.metrics.request_concurrency.successful.mean:.2f}",
+                    f"{benchmark.metrics.output_tokens_per_second.total.mean:.1f}",
+                    f"{benchmark.metrics.tokens_per_second.total.mean:.1f}",
                     (
-                        f"{benchmark.request_latency.successful.mean:.2f} / "
-                        f"{benchmark.request_latency.successful.median:.2f} / "
-                        f"{benchmark.request_latency.successful.percentiles.p99:.2f}"
+                        f"{benchmark.metrics.request_latency.successful.mean:.2f} / "
+                        f"{benchmark.metrics.request_latency.successful.median:.2f} / "
+                        f"{benchmark.metrics.request_latency.successful.percentiles.p99:.2f}"
                     ),
                     (
-                        f"{benchmark.time_to_first_token_ms.successful.mean:.1f} / "
-                        f"{benchmark.time_to_first_token_ms.successful.median:.1f} / "
-                        f"{benchmark.time_to_first_token_ms.successful.percentiles.p99:.1f}"
+                        f"{benchmark.metrics.time_to_first_token_ms.successful.mean:.1f} / "
+                        f"{benchmark.metrics.time_to_first_token_ms.successful.median:.1f} / "
+                        f"{benchmark.metrics.time_to_first_token_ms.successful.percentiles.p99:.1f}"
                     ),
                     (
-                        f"{benchmark.inter_token_latency_ms.successful.mean:.1f} / "
-                        f"{benchmark.inter_token_latency_ms.successful.median:.1f} / "
-                        f"{benchmark.inter_token_latency_ms.successful.percentiles.p99:.1f}"
+                        f"{benchmark.metrics.inter_token_latency_ms.successful.mean:.1f} / "
+                        f"{benchmark.metrics.inter_token_latency_ms.successful.median:.1f} / "
+                        f"{benchmark.metrics.inter_token_latency_ms.successful.percentiles.p99:.1f}"
                     ),
                     (
-                        f"{benchmark.time_per_output_token_ms.successful.mean:.1f} / "
-                        f"{benchmark.time_per_output_token_ms.successful.median:.1f} / "
-                        f"{benchmark.time_per_output_token_ms.successful.percentiles.p99:.1f}"
+                        f"{benchmark.metrics.time_per_output_token_ms.successful.mean:.1f} / "
+                        f"{benchmark.metrics.time_per_output_token_ms.successful.median:.1f} / "
+                        f"{benchmark.metrics.time_per_output_token_ms.successful.percentiles.p99:.1f}"
                     ),
                 ]
             )

src/guidellm/benchmark/progress.py

@@ -553,10 +553,10 @@ def handle_update_benchmark_compiled(
         progress_state.compiling = False
         progress_state.ended = True
         progress_state.requests_rate = (
-            current_benchmark.requests_per_second.successful.mean
+            current_benchmark.metric.requests_per_second.successful.mean
         )
         progress_state.requests_processing = (
-            current_benchmark.requests_concurrency.successful.mean
+            current_benchmark.metric.requests_concurrency.successful.mean
         )
 
     def handle_end(self, result: BenchmarkerResult):  # noqa: ARG002
@@ -647,22 +647,22 @@ def handle_update_benchmark_compiled(
         progress_state.requests_successful = current_benchmark.successful_total
         progress_state.requests_errored = current_benchmark.errored_total
         progress_state.output_tokens = (
-            current_benchmark.output_token_count.successful.mean
+            current_benchmark.metric.output_token_count.successful.mean
         )
         progress_state.prompt_tokens = (
-            current_benchmark.prompt_token_count.successful.mean
+            current_benchmark.metric.prompt_token_count.successful.mean
         )
         progress_state.output_tokens_rate = (
-            current_benchmark.output_tokens_per_second.successful.mean
+            current_benchmark.metric.output_tokens_per_second.successful.mean
         )
         progress_state.total_tokens_rate = (
-            current_benchmark.tokens_per_second.successful.mean
+            current_benchmark.metric.tokens_per_second.successful.mean
         )
         progress_state.tokens_ttft = (
-            current_benchmark.time_to_first_token_ms.successful.mean
+            current_benchmark.metric.time_to_first_token_ms.successful.mean
         )
         progress_state.tokens_itl = (
-            current_benchmark.inter_token_latency_ms.successful.mean
+            current_benchmark.metric.inter_token_latency_ms.successful.mean
         )
 
     def create_task_progress_state(