Merge branch 'main' into feat/multi-turn-tools-chat

Signed-off-by: Jared O'Connell <joconnel@redhat.com>

Author: jaredoconnell

AGENTS.md

@@ -65,13 +65,24 @@ tox -e tests -- -m regression
 - Use appropriate markers (`smoke`, `sanity`, `regression`)
 - Tests should be placed in files matching the name and path of the file under tests. E.g. `src/guidellm/benchmark/schemas/generative/entrypoints.py` -> `tests/unit/benchmark/schemas/generative/test_entrypoints.py`.
 
-### Style Requirements
+### Quality Requirements
 
 - All Python code must pass linting and formatting
 - All Python code must pass type checking
 - All tests must pass before committing
 - Markdown files must be properly formatted
+
+### Style Requirements
+
 - Public functions in `src/` code must use the reStructuredText docstring format
+- All imports **SHALL** be done at the top of the file
+- **DO NOT** use `getattr` or `setattr` as it hides incorrect usage of types
+
+### Design Requirements
+
+- Only touch sections of code that need to be changed for the given task
+- When handling variant-specific logic, encapsulate it in methods on registry class implementations rather than adding if/else branches to generic code paths
+- Class implementations must fully encapsulate their unique logic and that logic must not leak into caller code paths.
 
 ## Common Tasks
 

src/guidellm/backends/__init__.py

@@ -11,7 +11,7 @@
 
 from guidellm.extras.vllm import HAS_VLLM
 
-from .backend import Backend, BackendArgs, BackendType
+from .backend import Backend, BackendArgs
 from .openai import (
     AudioRequestHandler,
     ChatCompletionsRequestHandler,
@@ -32,7 +32,6 @@
     "AudioRequestHandler",
     "Backend",
     "BackendArgs",
-    "BackendType",
     "ChatCompletionsRequestHandler",
     "OpenAIHTTPBackend",
     "OpenAIRequestHandler",

src/guidellm/backends/backend.py

@@ -8,30 +8,62 @@
 
 from __future__ import annotations
 
-from abc import abstractmethod
-from typing import Literal
+from abc import ABC, abstractmethod
+from typing import ClassVar
 
-from pydantic import BaseModel, ConfigDict
+from pydantic import ConfigDict, Field
 
 from guidellm.scheduler import BackendInterface
-from guidellm.schemas import GenerationRequest, GenerationResponse
+from guidellm.schemas import (
+    GenerationRequest,
+    GenerationResponse,
+    PydanticClassRegistryMixin,
+)
 from guidellm.utils.registry import RegistryMixin
 
 __all__ = [
     "Backend",
     "BackendArgs",
-    "BackendType",
 ]
 
 
-BackendType = Literal["openai_http", "vllm_python"]
+class BackendArgs(PydanticClassRegistryMixin["BackendArgs"], ABC):
+    """
+    Base class for backend creation arguments.
+
+    This class serves as a base for defining argument models used in the creation
+    of backend instances. It inherits from PydanticClassRegistryMixin to enable
+    automatic registration of subclasses, allowing for flexible and extensible
+    backend configurations.
+
+    :cvar schema_discriminator: Field name for polymorphic deserialization
+    """
+
+    model_config = ConfigDict(
+        extra="forbid",
+        serialize_by_alias=True,
+        ser_json_bytes="base64",
+        val_json_bytes="base64",
+    )
+
+    schema_discriminator: ClassVar[str] = "type"
+
+    @classmethod
+    def __pydantic_schema_base_type__(cls) -> type[BackendArgs]:
+        """
+        Return base type for polymorphic validation hierarchy.
 
+        :return: Base BackendArgs class for schema validation
+        """
+        if cls.__name__ == "BackendArgs":
+            return cls
 
-class BackendArgs(BaseModel):
-    """Base class for backend creation argument models."""
+        return BackendArgs
 
-    # Allow for extra fields until we make BackendArgs the sole source of truth
-    model_config = ConfigDict(extra="allow")
+    type_: str = Field(
+        alias="type",
+        description="Type identifier for the backend configuration.",
+    )
 
 
 class Backend(
@@ -57,18 +89,19 @@ class Backend(
     ::
         @Backend.register("my_backend")
         class MyBackend(Backend):
-            def __init__(self, api_key: str):
-                super().__init__("my_backend")
-                self.api_key = api_key
+            def __init__(self, args: MyBackendArgs):
+                super().__init__(args)
+                self.api_key = args.api_key
 
             async def process_startup(self):
                 self.client = MyAPIClient(self.api_key)
 
-        backend = Backend.create("my_backend", api_key="secret")
+        args = MyBackendArgs(api_key="secret")
+        backend = Backend.create(args)
     """
 
     @classmethod
-    def create(cls, type_: str, **kwargs) -> Backend:
+    def create(cls, args: BackendArgs) -> Backend:
         """
         Create a backend instance based on the backend type.
 
@@ -77,6 +110,7 @@ def create(cls, type_: str, **kwargs) -> Backend:
         :return: An instance of a subclass of Backend
         :raises ValueError: If the backend type is not registered
         """
+        type_ = args.type_
 
         backend = cls.get_registered_object(type_)
 
@@ -86,34 +120,15 @@ def create(cls, type_: str, **kwargs) -> Backend:
                 f"Available types: {list(cls.registry.keys()) if cls.registry else []}"
             )
 
-        return backend(**kwargs)
-
-    @classmethod
-    def get_backend_args(cls, type_: str) -> type[BackendArgs]:
-        """
-        Return the Pydantic model class for the backend's creation arguments.
+        return backend(args)
 
-        :param type_: The backend type identifier
-        :return: The backend's BackendArgs subclass
-        :raises ValueError: If the backend type is not registered
-        """
-        backend_class = cls.get_registered_object(type_)
-
-        if backend_class is None:
-            raise ValueError(
-                f"Backend type '{type_}' is not registered. "
-                f"Available types: {list(cls.registry.keys()) if cls.registry else []}"
-            )
-
-        return backend_class.backend_args()
-
-    def __init__(self, type_: str):
+    def __init__(self, args: BackendArgs):
         """
         Initialize a backend instance.
 
         :param type_: The backend type identifier
         """
-        self.type_ = type_
+        self.type_ = args.type_
 
     @property
     def processes_limit(self) -> int | None:
@@ -130,19 +145,6 @@ def requests_limit(self) -> int | None:
         """
         return None
 
-    @classmethod
-    @abstractmethod
-    def backend_args(cls) -> type[BackendArgs]:
-        """
-        Return the Pydantic model class for this backend's creation arguments.
-
-        The model defines the parameters (e.g. target, model) that the CLI/benchmark
-        supply when creating the backend. Used for validation and error messages.
-
-        :return: A BackendArgs subclass whose fields are the creation params
-        """
-        ...
-
     @abstractmethod
     async def default_model(self) -> str:
         """