Merge branch 'main' into fix/worker-unusable-terminal-response

Author: ushaket

pyproject.toml

@@ -84,8 +84,8 @@ audio = [
     # Lowest version with full torchcodec support
     "datasets[audio]>=4.1.0",
     # Torchcodec needs specific torch version
-    "torch==2.9.*",
-    "torchcodec==0.8.*",
+    "torch==2.10.*",
+    "torchcodec==0.10.*",
 ]
 vision = [
     "datasets[vision]",

src/guidellm/benchmark/outputs/console.py

@@ -467,39 +467,39 @@ def print_server_throughput_table(self, report: GenerativeBenchmarksReport):
                 type_="text",
             )
             columns.add_stats(
-                benchmark.metrics.requests_per_second,
+                benchmark.metrics.request_concurrency,
                 status="total",
                 group="Requests",
-                name="Per Sec",
+                name="Concurrency",
                 types=("median", "mean"),
             )
             columns.add_stats(
-                benchmark.metrics.request_concurrency,
+                benchmark.metrics.requests_per_second,
                 status="total",
                 group="Requests",
-                name="Concurrency",
-                types=("median", "mean"),
+                name="Per Sec",
+                types=("mean",),
             )
             columns.add_stats(
                 benchmark.metrics.prompt_tokens_per_second,
                 status="total",
                 group="Input Tokens",
                 name="Per Sec",
-                types=("median", "mean"),
+                types=("mean",),
             )
             columns.add_stats(
                 benchmark.metrics.output_tokens_per_second,
                 status="total",
                 group="Output Tokens",
                 name="Per Sec",
-                types=("median", "mean"),
+                types=("mean",),
             )
             columns.add_stats(
                 benchmark.metrics.tokens_per_second,
                 status="total",
                 group="Total Tokens",
                 name="Per Sec",
-                types=("median", "mean"),
+                types=("mean",),
             )
 
         headers, values = columns.get_table_data()

src/guidellm/schemas/base.py

@@ -56,6 +56,8 @@ class ReloadableBaseModel(BaseModel):
         use_enum_values=True,
         from_attributes=True,
         arbitrary_types_allowed=True,
+        ser_json_bytes="base64",
+        val_json_bytes="base64",
     )
 
     @classmethod
@@ -169,6 +171,8 @@ class MyModel(StandardBaseModel):
         extra="ignore",
         use_enum_values=True,
         from_attributes=True,
+        ser_json_bytes="base64",
+        val_json_bytes="base64",
     )
 
     @classmethod
@@ -198,6 +202,8 @@ class StandardBaseDict(StandardBaseModel):
         use_enum_values=True,
         from_attributes=True,
         arbitrary_types_allowed=True,
+        ser_json_bytes="base64",
+        val_json_bytes="base64",
     )