Documentation pass

Author: sjmonson

src/guidellm/__main__.py

@@ -39,7 +39,11 @@ def cli():
         click.Choice(get_builtin_scenarios()),
     ),
     default=None,
-    help=("TODO: A scenario or path to config"),
+    help=(
+        "The name of a builtin scenario or path to a config file. "
+        "Missing values from the config will use defaults. "
+        "Options specified on the commandline will override the scenario."
+    ),
 )
 @click.option(
     "--target",
@@ -277,9 +281,10 @@ def benchmark(
             _scenario = GenerativeTextScenario.model_validate(overrides)
         elif isinstance(scenario, Path):
             _scenario = GenerativeTextScenario.from_file(scenario, overrides)
-        else:
+        else:  # Only builtins can make it here; click will catch anything else
             _scenario = GenerativeTextScenario.from_builtin(scenario, overrides)
     except ValidationError as e:
+        # Translate pydantic valdation error to click argument error
         errs = e.errors(include_url=False, include_context=True, include_input=True)
         param_name = "--" + str(errs[0]["loc"][0]).replace("_", "-")
         raise click.BadParameter(

src/guidellm/benchmark/scenario.py

@@ -21,10 +21,16 @@
 
 @cache
 def get_builtin_scenarios() -> list[str]:
+    """Returns list of builtin scenario names."""
     return [p.stem for p in SCENARIO_DIR.glob("*.json")]
 
 
 def parse_float_list(value: Union[str, float, list[float]]) -> list[float]:
+    """
+    Parse a comma separated string to a list of float
+    or convert single float list of one or pass float
+    list through.
+    """
     if isinstance(value, (int, float)):
         return [value]
     elif isinstance(value, list):
@@ -44,21 +50,30 @@ def parse_float_list(value: Union[str, float, list[float]]) -> list[float]:
 
 
 class Scenario(StandardBaseModel):
+    """
+    Parent Scenario class with common options for all benchmarking types.
+    """
+
     target: str
 
     @classmethod
     def from_builtin(cls: type[T], name: str, overrides: Optional[dict] = None) -> T:
         filename = SCENARIO_DIR / f"{name}.json"
 
         if not filename.is_file():
-            raise ValueError(f"{name} is not a vaild builtin scenario")
+            raise ValueError(f"{name} is not a valid builtin scenario")
 
         return cls.from_file(filename, overrides)
 
 
 class GenerativeTextScenario(Scenario):
-    # FIXME: This solves an issue with Pydantic and class types
+    """
+    Scenario class for generative text benchmarks.
+    """
+
     class Config:
+        # NOTE: This prevents errors due to unvalidatable
+        # types like PreTrainedTokenizerBase
         arbitrary_types_allowed = True
 
     backend_type: BackendType = "openai_http"

src/guidellm/objects/pydantic.py

@@ -10,6 +10,7 @@
 
 T = TypeVar("T", bound="StandardBaseModel")
 
+
 class StandardBaseModel(BaseModel):
     """
     A base class for Pydantic models throughout GuideLLM enabling standard
@@ -37,9 +38,11 @@ def get_default(cls: type[T], field: str) -> Any:
         return cls.model_fields[field].default
 
     @classmethod
-    def from_file(
-        cls: type[T], filename: Path, overrides: Optional[dict] = None
-    ) -> T:
+    def from_file(cls: type[T], filename: Path, overrides: Optional[dict] = None) -> T:
+        """
+        Attempt to create a new instance of the model using
+        data loaded from json or yaml file.
+        """
         try:
             with filename.open() as f:
                 if str(filename).endswith((".yaml", ".yml")):
@@ -48,7 +51,7 @@ def from_file(
                     data = json.load(f)
         except (json.JSONDecodeError, yaml.YAMLError) as e:
             logger.error(f"Failed to parse {filename} as type {cls.__name__}")
-            raise e
+            raise ValueError(f"Error when parsing file: {filename}") from e
 
         data.update(overrides)
         return cls.model_validate(data)