chores: Verbose error in pbar+json format (#41)

* Verbose error in pbar+indented json outputs.

Signed-off-by: Rashid Kaleem <230885705+arekay-nv@users.noreply.github.com>

Author: arekay-nv

src/inference_endpoint/commands/benchmark.py

@@ -129,6 +129,8 @@ def on_complete_hook(self, result: QueryResult):
         self.count += 1
         if result.error:
             self.errors.append(f"Sample {result.id}: {result.error}")
+            if self.pbar:
+                self.pbar.set_postfix(refresh=True, errors=len(self.errors))
         elif self.collect_responses:
             self.responses[result.id] = result.response_output
 

src/inference_endpoint/core/types.py

@@ -66,7 +66,7 @@ class Query(msgspec.Struct, kw_only=True):
     Example:
         >>> query = Query(
         ...     data={"prompt": "Hello", "model": "Qwen/Qwen3-8B", "max_tokens": 100},
-        ...     headers={"Authorization": "Bearer token123"}
+        ...     headers={"Authorization": "Bearer token123"},
         ... )
     """
 

src/inference_endpoint/load_generator/session.py

@@ -156,7 +156,12 @@ def _run_test(
                     # TODO: After Zhihan's MR is merged, grab the scheduler class and other LG init settings
                     # from the runtime settings object
                     with (Path(report_dir) / "runtime_settings.json").open("w") as f:
-                        f.write(orjson.dumps(rt_settings_data).decode("utf-8"))
+                        f.write(
+                            orjson.dumps(
+                                rt_settings_data,
+                                option=orjson.OPT_INDENT_2 | orjson.OPT_SORT_KEYS,
+                            ).decode("utf-8")
+                        )
 
                     if dump_events_csv:
                         reporter.dump_to_csv(Path(report_dir) / "events.csv")

src/inference_endpoint/metrics/reporter.py

@@ -408,7 +408,9 @@ def to_json(self, save_to: os.PathLike | None = None) -> str:
         d["qps"] = self.qps
         d["tps"] = self.tps
         d["e2e_sample_latency_sec"] = self.e2e_sample_latency_sec
-        json_str = orjson.dumps(d).decode("utf-8")
+        json_str = orjson.dumps(
+            d, option=orjson.OPT_INDENT_2 | orjson.OPT_SORT_KEYS
+        ).decode("utf-8")
         if save_to is not None:
             with Path(save_to).open("w") as f:
                 f.write(json_str)