latest updates and working state

Author: markurtz

src/guidellm/__main__.py

@@ -32,6 +32,19 @@
 
 import click
 
+try:
+    import uvloop
+
+    HAS_UVLOOP: Annotated[
+        bool, "Flag indicating if uvloop is available for event loop optimization"
+    ] = True
+except ImportError:
+    uvloop = None
+
+    HAS_UVLOOP: Annotated[
+        bool, "Flag indicating if uvloop is available for event loop optimization"
+    ] = False
+
 from guidellm.backend import BackendType
 from guidellm.benchmark import (
     GenerativeConsoleBenchmarkerProgress,
@@ -401,6 +414,8 @@ def run(
     Supports multiple backends, data sources, output formats, and constraint types
     for flexible benchmark configuration.
     """
+    if HAS_UVLOOP:
+        asyncio.set_event_loop_policy(uvloop.EventLoopPolicy())
     asyncio.run(
         benchmark_generative_text(
             target=target,

src/guidellm/benchmark/progress.py

@@ -802,7 +802,11 @@ def start(self, strategy: SchedulingStrategy):
     def update(
         self, aggregator_update: AggregatorState, scheduler_state: SchedulerState
     ):
-        self.progress = scheduler_state.remaining_fraction
+        self.progress = (
+            (1.0 - scheduler_state.remaining_fraction)
+            if scheduler_state.remaining_fraction is not None
+            else 0.0
+        )
         status: Literal["in_warmup", "in_progress", "in_cooldown"] | None = (
             "in_progress"  # Need to handle requests_in_* isn't in aggregator_update
         )

src/guidellm/scheduler/constraints.py

@@ -456,10 +456,8 @@ def __call__(
 
         create_exceeded = state.created_requests >= max_num
         processed_exceeded = state.processed_requests >= max_num
-        remaining_fraction = min(
-            max(0.0, 1.0 - state.processed_requests / float(max_num)), 1.0
-        )
-        remaining_requests = max(0, max_num - state.processed_requests)
+        remaining_requests = min(max(0, max_num - state.processed_requests), max_num)
+        remaining_fraction = remaining_requests / float(max_num)
 
         return SchedulerUpdateAction(
             request_queuing="stop" if create_exceeded else "continue",
@@ -577,6 +575,8 @@ def __call__(
         current_time = time.time()
         elapsed = current_time - state.start_time
         duration_exceeded = elapsed >= max_duration
+        remaining_duration = min(max(0.0, max_duration - elapsed), max_duration)
+        remaining_fraction = remaining_duration / float(max_duration)
 
         return SchedulerUpdateAction(
             request_queuing="stop" if duration_exceeded else "continue",
@@ -589,8 +589,8 @@ def __call__(
                 "current_time": current_time,
             },
             progress=SchedulerUpdateActionProgress(
-                remaining_fraction=max(0.0, 1.0 - elapsed / float(max_duration)),
-                remaining_duration=max(0.0, max_duration - elapsed),
+                remaining_fraction=remaining_fraction,
+                remaining_duration=remaining_duration,
             ),
         )
 

src/guidellm/scheduler/objects.py

@@ -122,6 +122,7 @@ def __pydantic_schema_base_type__(cls) -> type[MeasuredRequestTimings]:
     schema_discriminator: ClassVar[str] = "timings_type"
 
     timings_type: Literal["measured_request_timings"] = Field(
+        default="measured_request_timings",
         description="Type identifier for the timing measurement",
     )
     request_start: float | None = Field(
@@ -414,7 +415,7 @@ class SchedulerState(StandardBaseModel):
     )
     end_processing_constraints: dict[str, SchedulerUpdateAction] = Field(
         default_factory=dict,
-        description="Constraints that triggered processing termination",
+        description="Constraints that triggered process ing termination",
     )
     scheduler_constraints: dict[str, SchedulerUpdateAction] = Field(
         default_factory=dict,
@@ -429,7 +430,7 @@ class SchedulerState(StandardBaseModel):
             "Estimated fraction for the remaining progress of the run, if known"
         ),
     )
-    remaining_requests: int | None = Field(
+    remaining_requests: float | None = Field(
         default=None,
         description="Estimated number of requests remaining to be processed, if known",
     )
@@ -447,7 +448,8 @@ class SchedulerState(StandardBaseModel):
         default=0, description="Total number of requests queued for processing"
     )
     pending_requests: int = Field(
-        default=0, description="Number of requests currently pending processing"
+        default=0,
+        description="Total number of requests pending processing within a worker",
     )
     processing_requests: int = Field(
         default=0, description="Number of requests currently being processed"