add ability to pass in multiple request rates

Author: DaltheCow

setup.py

@@ -25,6 +25,7 @@ def _setup_long_description() -> Tuple[str, str]:
         'openai',
         'requests',
         'transformers',
+        'click'
     ],
     extras_require={
         'dev': [

src/guidellm/executor/__init__.py

@@ -3,7 +3,7 @@
     Profile,
     ProfileGenerationModes,
     ProfileGenerator,
-    SingleProfileGenerator,
+    MultiProfileGenerator,
     SweepProfileGenerator,
 )
 
@@ -12,6 +12,6 @@
     "ProfileGenerationModes",
     "Profile",
     "ProfileGenerator",
-    "SingleProfileGenerator",
+    "MultiProfileGenerator",
     "SweepProfileGenerator",
 ]

src/guidellm/executor/executor.py

@@ -14,7 +14,7 @@ def __init__(
         self,
         request_generator: RequestGenerator,
         backend: Backend,
-        profile_mode: Union[str, ProfileGenerationModes] = "single",
+        profile_mode: Union[str, ProfileGenerationModes] = "multi",
         profile_args: Optional[Dict[str, Any]] = None,
         max_requests: Optional[int] = None,
         max_duration: Optional[float] = None,

src/guidellm/executor/profile_generator.py

@@ -1,7 +1,7 @@
 from abc import ABC, abstractmethod
 from dataclasses import dataclass
 from enum import Enum
-from typing import Optional, Union
+from typing import List, Optional, Union
 
 import numpy
 
@@ -12,13 +12,13 @@
     "ProfileGenerationModes",
     "Profile",
     "ProfileGenerator",
-    "SingleProfileGenerator",
+    "MultiProfileGenerator",
     "SweepProfileGenerator",
 ]
 
 
 class ProfileGenerationModes(Enum):
-    SINGLE = "single"
+    MULTI = "multi"
     SWEEP = "sweep"
 
 
@@ -61,25 +61,24 @@ def next_profile(
         pass
 
 
-@ProfileGenerator.register_generator(ProfileGenerationModes.SINGLE)
-class SingleProfileGenerator(ProfileGenerator):
-    def __init__(self, rate: float, rate_type: str, **kwargs):
-        super().__init__(ProfileGenerationModes.SINGLE)
-        self._rate = rate
+@ProfileGenerator.register_generator(ProfileGenerationModes.MULTI)
+class MultiProfileGenerator(ProfileGenerator):
+    def __init__(self, rate: List[float], rate_type: str, **kwargs):
+        super().__init__(ProfileGenerationModes.MULTI)
+        self._rates = rate
+        self._rate_index = 0
         self._rate_type = rate_type
         self._generated = False
 
     def next_profile(
         self, current_report: TextGenerationBenchmarkReport
     ) -> Optional[Profile]:
-        if self._generated:
+        if self._rate_index >= len(self._rates):
             return None
 
-        self._generated = True
-
         if self._rate_type == "constant":
             return Profile(
-                load_gen_mode=LoadGenerationModes.CONSTANT, load_gen_rate=self._rate
+                load_gen_mode=LoadGenerationModes.CONSTANT, load_gen_rate=self._rates[self._rate_index]
             )
 
         if self._rate_type == "synchronous":
@@ -89,7 +88,7 @@ def next_profile(
 
         if self._rate_type == "poisson":
             return Profile(
-                load_gen_mode=LoadGenerationModes.POISSON, load_gen_rate=self._rate
+                load_gen_mode=LoadGenerationModes.POISSON, load_gen_rate=self._rates[self._rate_index]
             )
 
         raise ValueError(f"Invalid rate type: {self._rate_type}")
@@ -151,4 +150,4 @@ def next_profile(
                 load_gen_mode=LoadGenerationModes.CONSTANT, load_gen_rate=rate
             )
 
-        return None
+        return None

src/guidellm/main.py

@@ -46,8 +46,9 @@
 @click.option(
     "--rate",
     type=float,
-    default="1.0",
+    default=[1.0],
     help="Rate to use for constant and poisson rate types",
+    multiple=True,
 )
 @click.option(
     "--num-seconds",
@@ -106,10 +107,11 @@ def main(
         raise ValueError(f"Unknown data type: {data_type}")
 
     # Create executor
+    profile_mode = rate_type if rate_type == "sweep" else "multi"
     executor = Executor(
         request_generator=request_generator,
         backend=backend,
-        profile_mode=rate_type,
+        profile_mode=profile_mode,
         profile_args={"rate_type": rate_type, "rate": rate},
         max_requests=num_requests,
         max_duration=num_seconds,